From 96776405a17034dcfd53d3ddf5d142d34bdbb657 Mon Sep 17 00:00:00 2001 From: Diego Devesa Date: Fri, 11 Oct 2024 15:34:45 +0200 Subject: [PATCH 01/36] ggml : move more prints to the ggml log system (#9839) * ggml : move more prints to the ggml log system * show BLAS OpenMP warnings in all builds using debug print --- ggml/src/ggml-alloc.c | 34 +++++++++++++++++----------------- ggml/src/ggml-backend.cpp | 32 ++++++++++++++++---------------- ggml/src/ggml-blas.cpp | 8 ++++---- ggml/src/ggml-cuda.cu | 22 +++++++++++----------- 4 files changed, 48 insertions(+), 48 deletions(-) diff --git a/ggml/src/ggml-alloc.c b/ggml/src/ggml-alloc.c index 70187b9b6..28548fbbb 100644 --- a/ggml/src/ggml-alloc.c +++ b/ggml/src/ggml-alloc.c @@ -14,7 +14,7 @@ //#define GGML_ALLOCATOR_DEBUG -//#define AT_PRINTF(...) fprintf(stderr, __VA_ARGS__) +//#define AT_PRINTF(...) GGML_LOG_DEBUG(__VA_ARGS__) #define AT_PRINTF(...) @@ -89,7 +89,7 @@ void ggml_tallocr_alloc(struct ggml_tallocr * talloc, struct ggml_tensor * tenso size = GGML_PAD(size, talloc->alignment); if (talloc->offset + size > ggml_backend_buffer_get_size(talloc->buffer)) { - fprintf(stderr, "%s: not enough space in the buffer to allocate %s (needed %zu, available %zu)\n", + GGML_LOG_ERROR("%s: not enough space in the buffer to allocate %s (needed %zu, available %zu)\n", __func__, tensor->name, size, ggml_backend_buffer_get_size(talloc->buffer) - talloc->offset); GGML_ABORT("not enough space in the buffer"); } @@ -172,7 +172,7 @@ static size_t ggml_dyn_tallocr_alloc(struct ggml_dyn_tallocr * alloc, size_t siz best_fit_block = alloc->n_free_blocks - 1; } else { // this should never happen - fprintf(stderr, "%s: not enough space in the buffer to allocate %zu bytes, largest block available %zu bytes\n", + GGML_LOG_ERROR("%s: not enough space in the buffer to allocate %zu bytes, largest block available %zu bytes\n", __func__, size, max_avail); GGML_ABORT("not enough space in the buffer"); } @@ -209,16 +209,16 @@ static size_t ggml_dyn_tallocr_alloc(struct ggml_dyn_tallocr * alloc, size_t siz } } } - fprintf(stderr, "max_size = %.2f MB: tensors: ", cur_max / 1024.0 / 1024.0); + GGML_LOG_DEBUG("max_size = %.2f MB: tensors: ", cur_max / 1024.0 / 1024.0); for (int i = 0; i < 1024; i++) { if (alloc->allocated_tensors[i].tensor) { - fprintf(stderr, "%s [%zx-%zx] (%.2f MB) ", alloc->allocated_tensors[i].tensor->name, + GGML_LOG_DEBUG("%s [%zx-%zx] (%.2f MB) ", alloc->allocated_tensors[i].tensor->name, alloc->allocated_tensors[i].offset, alloc->allocated_tensors[i].offset + ggml_nbytes(alloc->allocated_tensors[i].tensor), ggml_nbytes(alloc->allocated_tensors[i].tensor) / 1024.0 / 1024.0); } } - fprintf(stderr, "\n"); + GGML_LOG_DEBUG("\n"); } #endif @@ -768,13 +768,13 @@ bool ggml_gallocr_reserve_n(ggml_gallocr_t galloc, struct ggml_cgraph * graph, c // even if there are no tensors allocated in this buffer, we still need to allocate it to initialize views if (new_size > cur_size || galloc->buffers[i] == NULL) { #ifndef NDEBUG - fprintf(stderr, "%s: reallocating %s buffer from size %.02f MiB to %.02f MiB\n", __func__, ggml_backend_buft_name(galloc->bufts[i]), cur_size / 1024.0 / 1024.0, new_size / 1024.0 / 1024.0); + GGML_LOG_DEBUG("%s: reallocating %s buffer from size %.02f MiB to %.02f MiB\n", __func__, ggml_backend_buft_name(galloc->bufts[i]), cur_size / 1024.0 / 1024.0, new_size / 1024.0 / 1024.0); #endif ggml_backend_buffer_free(galloc->buffers[i]); galloc->buffers[i] = ggml_backend_buft_alloc_buffer(galloc->bufts[i], new_size); if (galloc->buffers[i] == NULL) { - fprintf(stderr, "%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(galloc->bufts[i]), new_size); + GGML_LOG_ERROR("%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(galloc->bufts[i]), new_size); return false; } ggml_backend_buffer_set_usage(galloc->buffers[i], GGML_BACKEND_BUFFER_USAGE_COMPUTE); @@ -825,14 +825,14 @@ static bool ggml_gallocr_node_needs_realloc(ggml_gallocr_t galloc, struct ggml_t static bool ggml_gallocr_needs_realloc(ggml_gallocr_t galloc, struct ggml_cgraph * graph) { if (galloc->n_nodes != graph->n_nodes) { #ifndef NDEBUG - fprintf(stderr, "%s: graph has different number of nodes\n", __func__); + GGML_LOG_DEBUG("%s: graph has different number of nodes\n", __func__); #endif return true; } if (galloc->n_leafs != graph->n_leafs) { #ifndef NDEBUG - fprintf(stderr, "%s: graph has different number of leafs\n", __func__); + GGML_LOG_DEBUG("%s: graph has different number of leafs\n", __func__); #endif return true; } @@ -843,7 +843,7 @@ static bool ggml_gallocr_needs_realloc(ggml_gallocr_t galloc, struct ggml_cgraph if (!ggml_gallocr_node_needs_realloc(galloc, node, &node_alloc->dst)) { #ifndef NDEBUG - fprintf(stderr, "%s: node %s is not valid\n", __func__, node->name); + GGML_LOG_DEBUG("%s: node %s is not valid\n", __func__, node->name); #endif return true; } @@ -855,7 +855,7 @@ static bool ggml_gallocr_needs_realloc(ggml_gallocr_t galloc, struct ggml_cgraph } if (!ggml_gallocr_node_needs_realloc(galloc, src, &node_alloc->src[j])) { #ifndef NDEBUG - fprintf(stderr, "%s: src %d (%s) of node %s is not valid\n", __func__, j, src->name, node->name); + GGML_LOG_DEBUG("%s: src %d (%s) of node %s is not valid\n", __func__, j, src->name, node->name); #endif return true; } @@ -869,14 +869,14 @@ bool ggml_gallocr_alloc_graph(ggml_gallocr_t galloc, struct ggml_cgraph * graph) if (ggml_gallocr_needs_realloc(galloc, graph)) { if (galloc->n_buffers == 1) { #ifndef NDEBUG - fprintf(stderr, "%s: reallocating buffers automatically\n", __func__); + GGML_LOG_DEBUG("%s: reallocating buffers automatically\n", __func__); #endif if (!ggml_gallocr_reserve(galloc, graph)) { return false; } } else { #ifndef NDEBUG - fprintf(stderr, "%s: cannot reallocate multi buffer graph automatically, call reserve\n", __func__); + GGML_LOG_DEBUG("%s: cannot reallocate multi buffer graph automatically, call reserve\n", __func__); #endif return false; } @@ -940,7 +940,7 @@ static bool alloc_tensor_range(struct ggml_context * ctx, ggml_backend_buffer_t buffer = ggml_backend_buft_alloc_buffer(buft, size); if (buffer == NULL) { #ifndef NDEBUG - fprintf(stderr, "%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(buft), size); + GGML_LOG_DEBUG("%s: failed to allocate %s buffer of size %zu\n", __func__, ggml_backend_buft_name(buft), size); #endif for (size_t i = 0; i < *n_buffers; i++) { ggml_backend_buffer_free((*buffers)[i]); @@ -990,7 +990,7 @@ ggml_backend_buffer_t ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_conte } if (this_size > max_size) { - fprintf(stderr, "%s: tensor %s is too large to fit in a %s buffer (tensor size: %zu, max buffer size: %zu)\n", + GGML_LOG_ERROR("%s: tensor %s is too large to fit in a %s buffer (tensor size: %zu, max buffer size: %zu)\n", __func__, t->name, ggml_backend_buft_name(buft), this_size, max_size); @@ -1022,7 +1022,7 @@ ggml_backend_buffer_t ggml_backend_alloc_ctx_tensors_from_buft(struct ggml_conte if (n_buffers == 0) { #ifndef NDEBUG - fprintf(stderr, "%s: all tensors in the context are already allocated\n", __func__); + GGML_LOG_DEBUG("%s: all tensors in the context are already allocated\n", __func__); #endif return NULL; } diff --git a/ggml/src/ggml-backend.cpp b/ggml/src/ggml-backend.cpp index fb1d3ead3..15d650150 100644 --- a/ggml/src/ggml-backend.cpp +++ b/ggml/src/ggml-backend.cpp @@ -379,7 +379,7 @@ void ggml_backend_tensor_copy(struct ggml_tensor * src, struct ggml_tensor * dst ggml_backend_tensor_get(src, dst->data, 0, ggml_nbytes(src)); } else if (!ggml_backend_buffer_copy_tensor(src, dst)) { #ifndef NDEBUG - fprintf(stderr, "%s: warning: slow copy from %s to %s\n", __func__, ggml_backend_buffer_name(src->buffer), ggml_backend_buffer_name(dst->buffer)); + GGML_LOG_DEBUG("%s: warning: slow copy from %s to %s\n", __func__, ggml_backend_buffer_name(src->buffer), ggml_backend_buffer_name(dst->buffer)); #endif size_t nbytes = ggml_nbytes(src); void * data = malloc(nbytes); @@ -571,7 +571,7 @@ struct ggml_backend_registry { void register_backend(ggml_backend_reg_t reg) { #ifndef NDEBUG - fprintf(stderr, "%s: registered backend %s (%zu devices)\n", + GGML_LOG_DEBUG("%s: registered backend %s (%zu devices)\n", __func__, ggml_backend_reg_name(reg), ggml_backend_reg_dev_count(reg)); #endif backends.push_back(reg); @@ -582,7 +582,7 @@ struct ggml_backend_registry { void register_device(ggml_backend_dev_t device) { #ifndef NDEBUG - fprintf(stderr, "%s: registered device %s (%s)\n", __func__, ggml_backend_dev_name(device), ggml_backend_dev_description(device)); + GGML_LOG_DEBUG("%s: registered device %s (%s)\n", __func__, ggml_backend_dev_name(device), ggml_backend_dev_description(device)); #endif devices.push_back(device); } @@ -773,7 +773,7 @@ static ggml_backend_buffer_t ggml_backend_cpu_buffer_type_alloc_buffer(ggml_back size += TENSOR_ALIGNMENT; // malloc may return an address that is not aligned void * data = malloc(size); // TODO: use GGML_ALIGNED_MALLOC (move to ggml-impl.h) if (data == NULL) { - fprintf(stderr, "%s: failed to allocate buffer of size %zu\n", __func__, size); + GGML_LOG_ERROR("%s: failed to allocate buffer of size %zu\n", __func__, size); return NULL; } @@ -836,7 +836,7 @@ static ggml_backend_buffer_t ggml_backend_cpu_hbm_buffer_type_alloc_buffer(ggml_ void * ptr; int result = hbw_posix_memalign(&ptr, ggml_backend_cpu_buffer_type_get_alignment(buft), size); if (result != 0) { - fprintf(stderr, "failed to allocate HBM buffer of size %zu\n", size); + GGML_LOG_ERROR("failed to allocate HBM buffer of size %zu\n", size); return NULL; } @@ -1459,7 +1459,7 @@ static int ggml_backend_sched_backend_from_buffer(ggml_backend_sched_t sched, co } #ifndef NDEBUG - fprintf(stderr, "%s: warning: no backend supports op %s with a weight with buffer type %s used in tensor %s, the weight will need to be copied\n", + GGML_LOG_DEBUG("%s: warning: no backend supports op %s with a weight with buffer type %s used in tensor %s, the weight will need to be copied\n", __func__, ggml_op_desc(tensor), ggml_backend_buffer_name(buffer), tensor->name); #endif @@ -1548,13 +1548,13 @@ static void ggml_backend_sched_print_assignments(ggml_backend_sched_t sched, str for (int i = 0; i < graph->n_nodes; i++) { if (cur_split < sched->n_splits && i == sched->splits[cur_split].i_start) { ggml_backend_t split_backend = sched->backends[sched->splits[cur_split].backend_id]; - fprintf(stderr, "\n## SPLIT #%d: %s # %d inputs: ", cur_split, ggml_backend_name(split_backend), + GGML_LOG_DEBUG("\n## SPLIT #%d: %s # %d inputs: ", cur_split, ggml_backend_name(split_backend), sched->splits[cur_split].n_inputs); for (int j = 0; j < sched->splits[cur_split].n_inputs; j++) { - fprintf(stderr, "[%s (%5.5s)] ", sched->splits[cur_split].inputs[j]->name, + GGML_LOG_DEBUG("[%s (%5.5s)] ", sched->splits[cur_split].inputs[j]->name, fmt_size(ggml_nbytes(sched->splits[cur_split].inputs[j]))); } - fprintf(stderr, "\n"); + GGML_LOG_DEBUG("\n"); cur_split++; } struct ggml_tensor * node = graph->nodes[i]; @@ -1562,7 +1562,7 @@ static void ggml_backend_sched_print_assignments(ggml_backend_sched_t sched, str continue; } ggml_backend_t tensor_backend = ggml_backend_sched_get_tensor_backend(sched, node); - fprintf(stderr, "node #%3d (%10.10s): %20.20s (%5.5s) [%5.5s %8.8s]:", i, ggml_op_name(node->op), node->name, + GGML_LOG_DEBUG("node #%3d (%10.10s): %20.20s (%5.5s) [%5.5s %8.8s]:", i, ggml_op_name(node->op), node->name, fmt_size(ggml_nbytes(node)), tensor_backend ? ggml_backend_name(tensor_backend) : "NULL", GET_CAUSE(node)); for (int j = 0; j < GGML_MAX_SRC; j++) { struct ggml_tensor * src = node->src[j]; @@ -1570,10 +1570,10 @@ static void ggml_backend_sched_print_assignments(ggml_backend_sched_t sched, str continue; } ggml_backend_t src_backend = ggml_backend_sched_get_tensor_backend(sched, src); - fprintf(stderr, " %20.20s (%5.5s) [%5.5s %8.8s]", src->name, + GGML_LOG_DEBUG(" %20.20s (%5.5s) [%5.5s %8.8s]", src->name, fmt_size(ggml_nbytes(src)), src_backend ? ggml_backend_name(src_backend) : "NULL", GET_CAUSE(src)); } - fprintf(stderr, "\n"); + GGML_LOG_DEBUG("\n"); } } @@ -2087,11 +2087,11 @@ static bool ggml_backend_sched_alloc_splits(ggml_backend_sched_t sched) { // the re-allocation may cause the split inputs to be moved to a different address ggml_backend_sched_synchronize(sched); #ifndef NDEBUG - fprintf(stderr, "%s: failed to allocate graph, reserving (backend_ids_changed = %d)\n", __func__, backend_ids_changed); + GGML_LOG_DEBUG("%s: failed to allocate graph, reserving (backend_ids_changed = %d)\n", __func__, backend_ids_changed); #endif ggml_gallocr_reserve_n(sched->galloc, &sched->graph, sched->node_backend_ids, sched->leaf_backend_ids); if (!ggml_gallocr_alloc_graph(sched->galloc, &sched->graph)) { - fprintf(stderr, "%s: failed to allocate graph\n", __func__); + GGML_LOG_ERROR("%s: failed to allocate graph\n", __func__); return false; } } @@ -2485,7 +2485,7 @@ struct ggml_backend_graph_copy ggml_backend_graph_copy(ggml_backend_t backend, s struct ggml_context * ctx_unallocated = ggml_init(params); if (ctx_allocated == NULL || ctx_unallocated == NULL) { - fprintf(stderr, "failed to allocate context for graph copy\n"); + GGML_LOG_ERROR("%s: failed to allocate context for graph copy\n", __func__); ggml_hash_set_free(&hash_set); free(node_copies); free(node_init); @@ -2508,7 +2508,7 @@ struct ggml_backend_graph_copy ggml_backend_graph_copy(ggml_backend_t backend, s // allocate nodes ggml_backend_buffer_t buffer = ggml_backend_alloc_ctx_tensors(ctx_allocated, backend); if (buffer == NULL) { - fprintf(stderr, "failed to allocate buffer for graph copy\n"); + GGML_LOG_ERROR("%s: failed to allocate buffer for graph copy\n", __func__); ggml_hash_set_free(&hash_set); free(node_copies); free(node_init); diff --git a/ggml/src/ggml-blas.cpp b/ggml/src/ggml-blas.cpp index 55f724586..7875ec86d 100644 --- a/ggml/src/ggml-blas.cpp +++ b/ggml/src/ggml-blas.cpp @@ -297,14 +297,14 @@ ggml_backend_t ggml_backend_blas_init(void) { /* .context = */ ctx, }; -#if !defined(NDEBUG) && defined(OPENBLAS_VERSION) && defined(GGML_USE_OPENMP) +#if defined(OPENBLAS_VERSION) && defined(GGML_USE_OPENMP) if (openblas_get_parallel() != OPENBLAS_OPENMP) { - fprintf(stderr, "%s: warning: ggml is using OpenMP, but OpenBLAS was compiled without OpenMP support\n", __func__); + GGML_LOG_DEBUG("%s: warning: ggml is using OpenMP, but OpenBLAS was compiled without OpenMP support\n", __func__); } #endif -#if !defined(NDEBUG) && defined(BLIS_ENABLE_CBLAS) && defined(GGML_USE_OPENMP) && !defined(BLIS_ENABLE_OPENMP) - fprintf(stderr, "%s: warning: ggml is using OpenMP, but BLIS was compiled without OpenMP support\n", __func__); +#if defined(BLIS_ENABLE_CBLAS) && defined(GGML_USE_OPENMP) && !defined(BLIS_ENABLE_OPENMP) + GGML_LOG_DEBUG("%s: warning: ggml is using OpenMP, but BLIS was compiled without OpenMP support\n", __func__); #endif return backend; diff --git a/ggml/src/ggml-cuda.cu b/ggml/src/ggml-cuda.cu index edb61abdf..1338bd458 100644 --- a/ggml/src/ggml-cuda.cu +++ b/ggml/src/ggml-cuda.cu @@ -291,7 +291,7 @@ struct ggml_cuda_pool_leg : public ggml_cuda_pool { return; } } - GGML_LOG_WARN(GGML_CUDA_NAME " buffer pool full, increase MAX_CUDA_BUFFERS\n"); + GGML_LOG_DEBUG(GGML_CUDA_NAME " buffer pool full, increase MAX_CUDA_BUFFERS\n"); ggml_cuda_set_device(device); CUDA_CHECK(cudaFree(ptr)); pool_size -= size; @@ -980,7 +980,7 @@ static void * ggml_cuda_host_malloc(size_t size) { if (err != cudaSuccess) { // clear the error cudaGetLastError(); - GGML_LOG_WARN("%s: failed to allocate %.2f MiB of pinned memory: %s\n", __func__, + GGML_LOG_DEBUG("%s: failed to allocate %.2f MiB of pinned memory: %s\n", __func__, size / 1024.0 / 1024.0, cudaGetErrorString(err)); return nullptr; } @@ -2406,7 +2406,7 @@ static bool ggml_backend_cuda_cpy_tensor_async(ggml_backend_t backend_src, ggml_ if (cuda_ctx_src->device != buf_ctx_src->device || cuda_ctx_dst->device != buf_ctx_dst->device) { #ifndef NDEBUG - GGML_LOG_WARN("%s: backend and buffer devices do not match\n", __func__); + GGML_LOG_DEBUG("%s: backend and buffer devices do not match\n", __func__); #endif return false; } @@ -2524,7 +2524,7 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend, if (ggml_cuda_info().devices[cuda_ctx->device].cc < CC_AMPERE) { cuda_ctx->cuda_graph->disable_due_to_gpu_arch = true; #ifndef NDEBUG - GGML_LOG_WARN("%s: disabling CUDA graphs due to GPU architecture\n", __func__); + GGML_LOG_DEBUG("%s: disabling CUDA graphs due to GPU architecture\n", __func__); #endif } } @@ -2575,14 +2575,14 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend, if (node->src[0] && node->src[0]->buffer && ggml_backend_buffer_is_cuda_split(node->src[0]->buffer)) { use_cuda_graph = false; // Split buffers are not supported by CUDA graph capture #ifndef NDEBUG - GGML_LOG_WARN("%s: disabling CUDA graphs due to split buffer\n", __func__); + GGML_LOG_DEBUG("%s: disabling CUDA graphs due to split buffer\n", __func__); #endif } if (node->op == GGML_OP_MUL_MAT_ID) { use_cuda_graph = false; // This node type is not supported by CUDA graph capture #ifndef NDEBUG - GGML_LOG_WARN("%s: disabling CUDA graphs due to mul_mat_id\n", __func__); + GGML_LOG_DEBUG("%s: disabling CUDA graphs due to mul_mat_id\n", __func__); #endif } @@ -2591,7 +2591,7 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend, // Changes in batch size or context size can cause changes to the grid size of some kernels. use_cuda_graph = false; #ifndef NDEBUG - GGML_LOG_WARN("%s: disabling CUDA graphs due to batch size > 1 [%s] [%ld %ld %ld %ld]\n", __func__, node->name, node->ne[0], node->ne[1], node->ne[2], node->ne[3]); + GGML_LOG_DEBUG("%s: disabling CUDA graphs due to batch size > 1 [%s] [%ld %ld %ld %ld]\n", __func__, node->name, node->ne[0], node->ne[1], node->ne[2], node->ne[3]); #endif } @@ -2603,7 +2603,7 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend, if (!ptr) { use_cuda_graph = false; #ifndef NDEBUG - GGML_LOG_WARN("%s: disabling CUDA graphs due to unsupported copy op\n", __func__); + GGML_LOG_DEBUG("%s: disabling CUDA graphs due to unsupported copy op\n", __func__); #endif } else { if (std::find(ggml_cuda_cpy_fn_ptrs.begin(), ggml_cuda_cpy_fn_ptrs.end(), ptr) == ggml_cuda_cpy_fn_ptrs.end()) { @@ -2627,7 +2627,7 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend, if (cuda_ctx->cuda_graph->number_consecutive_updates >= 4) { cuda_ctx->cuda_graph->disable_due_to_too_many_updates = true; #ifndef NDEBUG - GGML_LOG_WARN("%s: disabling CUDA graphs due to too many consecutive updates\n", __func__); + GGML_LOG_DEBUG("%s: disabling CUDA graphs due to too many consecutive updates\n", __func__); #endif } } @@ -2685,7 +2685,7 @@ static enum ggml_status ggml_backend_cuda_graph_compute(ggml_backend_t backend, use_cuda_graph = false; cuda_ctx->cuda_graph->disable_due_to_failed_graph_capture = true; #ifndef NDEBUG - GGML_LOG_WARN("%s: disabling CUDA graphs due to failed graph capture\n", __func__); + GGML_LOG_DEBUG("%s: disabling CUDA graphs due to failed graph capture\n", __func__); #endif } else { graph_evaluated_or_captured = true; // CUDA graph has been captured @@ -2854,7 +2854,7 @@ bool ggml_backend_cuda_register_host_buffer(void * buffer, size_t size) { // clear the error cudaGetLastError(); - GGML_LOG_WARN("%s: failed to register %.2f MiB of pinned memory: %s\n", __func__, + GGML_LOG_DEBUG("%s: failed to register %.2f MiB of pinned memory: %s\n", __func__, size / 1024.0 / 1024.0, cudaGetErrorString(err)); return false; } From 943d20b4111c746bcd9dbc7e4771de313b08b50c Mon Sep 17 00:00:00 2001 From: R0CKSTAR Date: Sat, 12 Oct 2024 13:09:53 +0800 Subject: [PATCH 02/36] musa : update doc (#9856) Signed-off-by: Xiaodong Ye --- README.md | 4 ++-- docs/build.md | 8 ++++++++ 2 files changed, 10 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 41e5e5448..dd4927b04 100644 --- a/README.md +++ b/README.md @@ -31,7 +31,7 @@ variety of hardware - locally and in the cloud. - Apple silicon is a first-class citizen - optimized via ARM NEON, Accelerate and Metal frameworks - AVX, AVX2 and AVX512 support for x86 architectures - 1.5-bit, 2-bit, 3-bit, 4-bit, 5-bit, 6-bit, and 8-bit integer quantization for faster inference and reduced memory use -- Custom CUDA kernels for running LLMs on NVIDIA GPUs (support for AMD GPUs via HIP) +- Custom CUDA kernels for running LLMs on NVIDIA GPUs (support for AMD GPUs via HIP and Moore Threads MTT GPUs via MUSA) - Vulkan and SYCL backend support - CPU+GPU hybrid inference to partially accelerate models larger than the total VRAM capacity @@ -413,7 +413,7 @@ Please refer to [Build llama.cpp locally](./docs/build.md) | [BLAS](./docs/build.md#blas-build) | All | | [BLIS](./docs/backend/BLIS.md) | All | | [SYCL](./docs/backend/SYCL.md) | Intel and Nvidia GPU | -| [MUSA](./docs/build.md#musa) | Moore Threads GPU | +| [MUSA](./docs/build.md#musa) | Moore Threads MTT GPU | | [CUDA](./docs/build.md#cuda) | Nvidia GPU | | [hipBLAS](./docs/build.md#hipblas) | AMD GPU | | [Vulkan](./docs/build.md#vulkan) | GPU | diff --git a/docs/build.md b/docs/build.md index faa0ecfa4..4e362ebc7 100644 --- a/docs/build.md +++ b/docs/build.md @@ -198,6 +198,8 @@ The following compilation options are also available to tweak performance: ### MUSA +This provides GPU acceleration using the MUSA cores of your Moore Threads MTT GPU. Make sure to have the MUSA SDK installed. You can download it from here: [MUSA SDK](https://developer.mthreads.com/sdk/download/musa). + - Using `make`: ```bash make GGML_MUSA=1 @@ -209,6 +211,12 @@ The following compilation options are also available to tweak performance: cmake --build build --config Release ``` +The environment variable [`MUSA_VISIBLE_DEVICES`](https://docs.mthreads.com/musa-sdk/musa-sdk-doc-online/programming_guide/Z%E9%99%84%E5%BD%95/) can be used to specify which GPU(s) will be used. + +The environment variable `GGML_CUDA_ENABLE_UNIFIED_MEMORY=1` can be used to enable unified memory in Linux. This allows swapping to system RAM instead of crashing when the GPU VRAM is exhausted. + +Most of the compilation options available for CUDA should also be available for MUSA, though they haven't been thoroughly tested yet. + ### hipBLAS This provides BLAS acceleration on HIP-supported AMD GPUs. From 11ac9800aff532715a5bc7991062c68ba3472e6e Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Sat, 12 Oct 2024 08:21:51 +0300 Subject: [PATCH 03/36] llama : improve infill support and special token detection (#9798) * llama : improve infill support ggml-ci * llama : add more FIM token strings ggml-ci * server : update prompt on slot restore (#9800) * gguf : deprecate old FIM token KVs --- common/arg.cpp | 248 +++++++++----------- common/common.cpp | 18 +- common/common.h | 19 +- examples/infill/infill.cpp | 14 +- examples/server/README.md | 2 +- examples/server/server.cpp | 150 ++++++------ gguf-py/gguf/constants.py | 26 ++- gguf-py/gguf/gguf_writer.py | 9 - include/llama.h | 17 +- src/llama-vocab.cpp | 38 ++- src/llama-vocab.h | 35 ++- src/llama.cpp | 452 ++++++++++++++++++++++-------------- 12 files changed, 601 insertions(+), 427 deletions(-) diff --git a/common/arg.cpp b/common/arg.cpp index 6014f5d8a..c4229a3a4 100644 --- a/common/arg.cpp +++ b/common/arg.cpp @@ -119,32 +119,6 @@ std::string common_arg::to_string() { // utils // -#ifdef __GNUC__ -#ifdef __MINGW32__ -#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(gnu_printf, __VA_ARGS__))) -#else -#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__))) -#endif -#else -#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) -#endif - -LLAMA_COMMON_ATTRIBUTE_FORMAT(1, 2) -static std::string format(const char * fmt, ...) { - va_list ap; - va_list ap2; - va_start(ap, fmt); - va_copy(ap2, ap); - int size = vsnprintf(NULL, 0, fmt, ap); - GGML_ASSERT(size >= 0 && size < INT_MAX); // NOLINT - std::vector buf(size + 1); - int size2 = vsnprintf(buf.data(), size + 1, fmt, ap2); - GGML_ASSERT(size2 == size); - va_end(ap2); - va_end(ap); - return std::string(buf.data(), size); -} - static void common_params_handle_model_default(common_params & params) { if (!params.hf_repo.empty()) { // short-hand to avoid specifying --hf-file -> default it to --model @@ -199,7 +173,7 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context continue; } } catch (std::exception & e) { - throw std::invalid_argument(format( + throw std::invalid_argument(string_format( "error while handling environment variable \"%s\": %s\n\n", opt.env, e.what())); } } @@ -220,7 +194,7 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context std::replace(arg.begin(), arg.end(), '_', '-'); } if (arg_to_options.find(arg) == arg_to_options.end()) { - throw std::invalid_argument(format("error: invalid argument: %s", arg.c_str())); + throw std::invalid_argument(string_format("error: invalid argument: %s", arg.c_str())); } auto opt = *arg_to_options[arg]; if (opt.has_value_from_env()) { @@ -252,7 +226,7 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context continue; } } catch (std::exception & e) { - throw std::invalid_argument(format( + throw std::invalid_argument(string_format( "error while handling argument \"%s\": %s\n\n" "usage:\n%s\n\nto show complete usage, run with -h", arg.c_str(), e.what(), arg_to_options[arg]->to_string().c_str())); @@ -391,28 +365,28 @@ common_params_context common_params_parser_init(common_params & params, llama_ex )); add_opt(common_arg( {"--verbose-prompt"}, - format("print a verbose prompt before generation (default: %s)", params.verbose_prompt ? "true" : "false"), + string_format("print a verbose prompt before generation (default: %s)", params.verbose_prompt ? "true" : "false"), [](common_params & params) { params.verbose_prompt = true; } )); add_opt(common_arg( {"--no-display-prompt"}, - format("don't print prompt at generation (default: %s)", !params.display_prompt ? "true" : "false"), + string_format("don't print prompt at generation (default: %s)", !params.display_prompt ? "true" : "false"), [](common_params & params) { params.display_prompt = false; } ).set_examples({LLAMA_EXAMPLE_MAIN})); add_opt(common_arg( {"-co", "--color"}, - format("colorise output to distinguish prompt and user input from generations (default: %s)", params.use_color ? "true" : "false"), + string_format("colorise output to distinguish prompt and user input from generations (default: %s)", params.use_color ? "true" : "false"), [](common_params & params) { params.use_color = true; } ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_INFILL, LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP})); add_opt(common_arg( {"-t", "--threads"}, "N", - format("number of threads to use during generation (default: %d)", params.cpuparams.n_threads), + string_format("number of threads to use during generation (default: %d)", params.cpuparams.n_threads), [](common_params & params, int value) { params.cpuparams.n_threads = value; if (params.cpuparams.n_threads <= 0) { @@ -472,14 +446,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex )); add_opt(common_arg( {"--cpu-strict"}, "<0|1>", - format("use strict CPU placement (default: %u)\n", (unsigned) params.cpuparams.strict_cpu), + string_format("use strict CPU placement (default: %u)\n", (unsigned) params.cpuparams.strict_cpu), [](common_params & params, const std::string & value) { params.cpuparams.strict_cpu = std::stoul(value); } )); add_opt(common_arg( {"--prio"}, "N", - format("set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.cpuparams.priority), + string_format("set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.cpuparams.priority), [](common_params & params, int prio) { if (prio < 0 || prio > 3) { throw std::invalid_argument("invalid value"); @@ -489,7 +463,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex )); add_opt(common_arg( {"--poll"}, "<0...100>", - format("use polling level to wait for work (0 - no polling, default: %u)\n", (unsigned) params.cpuparams.poll), + string_format("use polling level to wait for work (0 - no polling, default: %u)\n", (unsigned) params.cpuparams.poll), [](common_params & params, const std::string & value) { params.cpuparams.poll = std::stoul(value); } @@ -523,7 +497,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex )); add_opt(common_arg( {"--prio-batch"}, "N", - format("set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.cpuparams_batch.priority), + string_format("set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.cpuparams_batch.priority), [](common_params & params, int prio) { if (prio < 0 || prio > 3) { throw std::invalid_argument("invalid value"); @@ -567,7 +541,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_SPECULATIVE})); add_opt(common_arg( {"--prio-draft"}, "N", - format("set draft process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.draft_cpuparams.priority), + string_format("set draft process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.draft_cpuparams.priority), [](common_params & params, int prio) { if (prio < 0 || prio > 3) { throw std::invalid_argument("invalid value"); @@ -611,7 +585,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_SPECULATIVE})); add_opt(common_arg( {"--prio-batch-draft"}, "N", - format("set draft process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.draft_cpuparams_batch.priority), + string_format("set draft process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.draft_cpuparams_batch.priority), [](common_params & params, int prio) { if (prio < 0 || prio > 3) { throw std::invalid_argument("invalid value"); @@ -628,14 +602,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_SPECULATIVE})); add_opt(common_arg( {"--draft"}, "N", - format("number of tokens to draft for speculative decoding (default: %d)", params.n_draft), + string_format("number of tokens to draft for speculative decoding (default: %d)", params.n_draft), [](common_params & params, int value) { params.n_draft = value; } ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP})); add_opt(common_arg( {"-ps", "--p-split"}, "N", - format("speculative decoding split probability (default: %.1f)", (double)params.p_split), + string_format("speculative decoding split probability (default: %.1f)", (double)params.p_split), [](common_params & params, const std::string & value) { params.p_split = std::stof(value); } @@ -656,56 +630,56 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_LOOKUP})); add_opt(common_arg( {"-c", "--ctx-size"}, "N", - format("size of the prompt context (default: %d, 0 = loaded from model)", params.n_ctx), + string_format("size of the prompt context (default: %d, 0 = loaded from model)", params.n_ctx), [](common_params & params, int value) { params.n_ctx = value; } ).set_env("LLAMA_ARG_CTX_SIZE")); add_opt(common_arg( {"-n", "--predict", "--n-predict"}, "N", - format("number of tokens to predict (default: %d, -1 = infinity, -2 = until context filled)", params.n_predict), + string_format("number of tokens to predict (default: %d, -1 = infinity, -2 = until context filled)", params.n_predict), [](common_params & params, int value) { params.n_predict = value; } ).set_env("LLAMA_ARG_N_PREDICT")); add_opt(common_arg( {"-b", "--batch-size"}, "N", - format("logical maximum batch size (default: %d)", params.n_batch), + string_format("logical maximum batch size (default: %d)", params.n_batch), [](common_params & params, int value) { params.n_batch = value; } ).set_env("LLAMA_ARG_BATCH")); add_opt(common_arg( {"-ub", "--ubatch-size"}, "N", - format("physical maximum batch size (default: %d)", params.n_ubatch), + string_format("physical maximum batch size (default: %d)", params.n_ubatch), [](common_params & params, int value) { params.n_ubatch = value; } ).set_env("LLAMA_ARG_UBATCH")); add_opt(common_arg( {"--keep"}, "N", - format("number of tokens to keep from the initial prompt (default: %d, -1 = all)", params.n_keep), + string_format("number of tokens to keep from the initial prompt (default: %d, -1 = all)", params.n_keep), [](common_params & params, int value) { params.n_keep = value; } )); add_opt(common_arg( {"--no-context-shift"}, - format("disables context shift on inifinite text generation (default: %s)", params.ctx_shift ? "disabled" : "enabled"), + string_format("disables context shift on inifinite text generation (default: %s)", params.ctx_shift ? "disabled" : "enabled"), [](common_params & params) { params.ctx_shift = false; } ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_NO_CONTEXT_SHIFT")); add_opt(common_arg( {"--chunks"}, "N", - format("max number of chunks to process (default: %d, -1 = all)", params.n_chunks), + string_format("max number of chunks to process (default: %d, -1 = all)", params.n_chunks), [](common_params & params, int value) { params.n_chunks = value; } ).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_PERPLEXITY, LLAMA_EXAMPLE_RETRIEVAL})); add_opt(common_arg( {"-fa", "--flash-attn"}, - format("enable Flash Attention (default: %s)", params.flash_attn ? "enabled" : "disabled"), + string_format("enable Flash Attention (default: %s)", params.flash_attn ? "enabled" : "disabled"), [](common_params & params) { params.flash_attn = true; } @@ -721,7 +695,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex )); add_opt(common_arg( {"--no-perf"}, - format("disable internal libllama performance timings (default: %s)", params.no_perf ? "true" : "false"), + string_format("disable internal libllama performance timings (default: %s)", params.no_perf ? "true" : "false"), [](common_params & params) { params.no_perf = true; params.sparams.no_perf = true; @@ -733,7 +707,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex [](common_params & params, const std::string & value) { std::ifstream file(value); if (!file) { - throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str())); + throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str())); } // store the external file name in params params.prompt_file = value; @@ -749,7 +723,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex [](common_params & params, const std::string & value) { std::ifstream file(value); if (!file) { - throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str())); + throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str())); } params.in_files.push_back(value); } @@ -760,7 +734,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex [](common_params & params, const std::string & value) { std::ifstream file(value, std::ios::binary); if (!file) { - throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str())); + throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str())); } // store the external file name in params params.prompt_file = value; @@ -772,7 +746,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex )); add_opt(common_arg( {"-e", "--escape"}, - format("process escapes sequences (\\n, \\r, \\t, \\', \\\", \\\\) (default: %s)", params.escape ? "true" : "false"), + string_format("process escapes sequences (\\n, \\r, \\t, \\', \\\", \\\\) (default: %s)", params.escape ? "true" : "false"), [](common_params & params) { params.escape = true; } @@ -786,7 +760,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex )); add_opt(common_arg( {"-ptc", "--print-token-count"}, "N", - format("print token count every N tokens (default: %d)", params.n_print), + string_format("print token count every N tokens (default: %d)", params.n_print), [](common_params & params, int value) { params.n_print = value; } @@ -821,14 +795,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_MAIN})); add_opt(common_arg( {"-sp", "--special"}, - format("special tokens output enabled (default: %s)", params.special ? "true" : "false"), + string_format("special tokens output enabled (default: %s)", params.special ? "true" : "false"), [](common_params & params) { params.special = true; } ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER})); add_opt(common_arg( {"-cnv", "--conversation"}, - format( + string_format( "run in conversation mode:\n" "- does not print special tokens and suffix/prefix\n" "- interactive mode is also enabled\n" @@ -841,14 +815,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_MAIN})); add_opt(common_arg( {"-i", "--interactive"}, - format("run in interactive mode (default: %s)", params.interactive ? "true" : "false"), + string_format("run in interactive mode (default: %s)", params.interactive ? "true" : "false"), [](common_params & params) { params.interactive = true; } ).set_examples({LLAMA_EXAMPLE_MAIN})); add_opt(common_arg( {"-if", "--interactive-first"}, - format("run in interactive mode and wait for input right away (default: %s)", params.interactive_first ? "true" : "false"), + string_format("run in interactive mode and wait for input right away (default: %s)", params.interactive_first ? "true" : "false"), [](common_params & params) { params.interactive_first = true; } @@ -893,7 +867,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_MAIN})); add_opt(common_arg( {"--spm-infill"}, - format( + string_format( "use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: %s)", params.spm_infill ? "enabled" : "disabled" ), @@ -903,7 +877,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_INFILL})); add_opt(common_arg( {"--samplers"}, "SAMPLERS", - format("samplers that will be used for generation in the order, separated by \';\'\n(default: %s)", sampler_type_names.c_str()), + string_format("samplers that will be used for generation in the order, separated by \';\'\n(default: %s)", sampler_type_names.c_str()), [](common_params & params, const std::string & value) { const auto sampler_names = string_split(value, ';'); params.sparams.samplers = common_sampler_types_from_names(sampler_names, true); @@ -911,14 +885,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_sparam()); add_opt(common_arg( {"-s", "--seed"}, "SEED", - format("RNG seed (default: %d, use random seed for %d)", params.sparams.seed, LLAMA_DEFAULT_SEED), + string_format("RNG seed (default: %d, use random seed for %d)", params.sparams.seed, LLAMA_DEFAULT_SEED), [](common_params & params, const std::string & value) { params.sparams.seed = std::stoul(value); } ).set_sparam()); add_opt(common_arg( {"--sampling-seq"}, "SEQUENCE", - format("simplified sequence for samplers that will be used (default: %s)", sampler_type_chars.c_str()), + string_format("simplified sequence for samplers that will be used (default: %s)", sampler_type_chars.c_str()), [](common_params & params, const std::string & value) { params.sparams.samplers = common_sampler_types_from_chars(value); } @@ -932,14 +906,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_sparam()); add_opt(common_arg( {"--penalize-nl"}, - format("penalize newline tokens (default: %s)", params.sparams.penalize_nl ? "true" : "false"), + string_format("penalize newline tokens (default: %s)", params.sparams.penalize_nl ? "true" : "false"), [](common_params & params) { params.sparams.penalize_nl = true; } ).set_sparam()); add_opt(common_arg( {"--temp"}, "N", - format("temperature (default: %.1f)", (double)params.sparams.temp), + string_format("temperature (default: %.1f)", (double)params.sparams.temp), [](common_params & params, const std::string & value) { params.sparams.temp = std::stof(value); params.sparams.temp = std::max(params.sparams.temp, 0.0f); @@ -947,42 +921,42 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_sparam()); add_opt(common_arg( {"--top-k"}, "N", - format("top-k sampling (default: %d, 0 = disabled)", params.sparams.top_k), + string_format("top-k sampling (default: %d, 0 = disabled)", params.sparams.top_k), [](common_params & params, int value) { params.sparams.top_k = value; } ).set_sparam()); add_opt(common_arg( {"--top-p"}, "N", - format("top-p sampling (default: %.1f, 1.0 = disabled)", (double)params.sparams.top_p), + string_format("top-p sampling (default: %.1f, 1.0 = disabled)", (double)params.sparams.top_p), [](common_params & params, const std::string & value) { params.sparams.top_p = std::stof(value); } ).set_sparam()); add_opt(common_arg( {"--min-p"}, "N", - format("min-p sampling (default: %.1f, 0.0 = disabled)", (double)params.sparams.min_p), + string_format("min-p sampling (default: %.1f, 0.0 = disabled)", (double)params.sparams.min_p), [](common_params & params, const std::string & value) { params.sparams.min_p = std::stof(value); } ).set_sparam()); add_opt(common_arg( {"--tfs"}, "N", - format("tail free sampling, parameter z (default: %.1f, 1.0 = disabled)", (double)params.sparams.tfs_z), + string_format("tail free sampling, parameter z (default: %.1f, 1.0 = disabled)", (double)params.sparams.tfs_z), [](common_params & params, const std::string & value) { params.sparams.tfs_z = std::stof(value); } ).set_sparam()); add_opt(common_arg( {"--typical"}, "N", - format("locally typical sampling, parameter p (default: %.1f, 1.0 = disabled)", (double)params.sparams.typ_p), + string_format("locally typical sampling, parameter p (default: %.1f, 1.0 = disabled)", (double)params.sparams.typ_p), [](common_params & params, const std::string & value) { params.sparams.typ_p = std::stof(value); } ).set_sparam()); add_opt(common_arg( {"--repeat-last-n"}, "N", - format("last n tokens to consider for penalize (default: %d, 0 = disabled, -1 = ctx_size)", params.sparams.penalty_last_n), + string_format("last n tokens to consider for penalize (default: %d, 0 = disabled, -1 = ctx_size)", params.sparams.penalty_last_n), [](common_params & params, int value) { params.sparams.penalty_last_n = value; params.sparams.n_prev = std::max(params.sparams.n_prev, params.sparams.penalty_last_n); @@ -990,42 +964,42 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_sparam()); add_opt(common_arg( {"--repeat-penalty"}, "N", - format("penalize repeat sequence of tokens (default: %.1f, 1.0 = disabled)", (double)params.sparams.penalty_repeat), + string_format("penalize repeat sequence of tokens (default: %.1f, 1.0 = disabled)", (double)params.sparams.penalty_repeat), [](common_params & params, const std::string & value) { params.sparams.penalty_repeat = std::stof(value); } ).set_sparam()); add_opt(common_arg( {"--presence-penalty"}, "N", - format("repeat alpha presence penalty (default: %.1f, 0.0 = disabled)", (double)params.sparams.penalty_present), + string_format("repeat alpha presence penalty (default: %.1f, 0.0 = disabled)", (double)params.sparams.penalty_present), [](common_params & params, const std::string & value) { params.sparams.penalty_present = std::stof(value); } ).set_sparam()); add_opt(common_arg( {"--frequency-penalty"}, "N", - format("repeat alpha frequency penalty (default: %.1f, 0.0 = disabled)", (double)params.sparams.penalty_freq), + string_format("repeat alpha frequency penalty (default: %.1f, 0.0 = disabled)", (double)params.sparams.penalty_freq), [](common_params & params, const std::string & value) { params.sparams.penalty_freq = std::stof(value); } ).set_sparam()); add_opt(common_arg( {"--dynatemp-range"}, "N", - format("dynamic temperature range (default: %.1f, 0.0 = disabled)", (double)params.sparams.dynatemp_range), + string_format("dynamic temperature range (default: %.1f, 0.0 = disabled)", (double)params.sparams.dynatemp_range), [](common_params & params, const std::string & value) { params.sparams.dynatemp_range = std::stof(value); } ).set_sparam()); add_opt(common_arg( {"--dynatemp-exp"}, "N", - format("dynamic temperature exponent (default: %.1f)", (double)params.sparams.dynatemp_exponent), + string_format("dynamic temperature exponent (default: %.1f)", (double)params.sparams.dynatemp_exponent), [](common_params & params, const std::string & value) { params.sparams.dynatemp_exponent = std::stof(value); } ).set_sparam()); add_opt(common_arg( {"--mirostat"}, "N", - format("use Mirostat sampling.\nTop K, Nucleus, Tail Free and Locally Typical samplers are ignored if used.\n" + string_format("use Mirostat sampling.\nTop K, Nucleus, Tail Free and Locally Typical samplers are ignored if used.\n" "(default: %d, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)", params.sparams.mirostat), [](common_params & params, int value) { params.sparams.mirostat = value; @@ -1033,14 +1007,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_sparam()); add_opt(common_arg( {"--mirostat-lr"}, "N", - format("Mirostat learning rate, parameter eta (default: %.1f)", (double)params.sparams.mirostat_eta), + string_format("Mirostat learning rate, parameter eta (default: %.1f)", (double)params.sparams.mirostat_eta), [](common_params & params, const std::string & value) { params.sparams.mirostat_eta = std::stof(value); } ).set_sparam()); add_opt(common_arg( {"--mirostat-ent"}, "N", - format("Mirostat target entropy, parameter tau (default: %.1f)", (double)params.sparams.mirostat_tau), + string_format("Mirostat target entropy, parameter tau (default: %.1f)", (double)params.sparams.mirostat_tau), [](common_params & params, const std::string & value) { params.sparams.mirostat_tau = std::stof(value); } @@ -1069,7 +1043,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_sparam()); add_opt(common_arg( {"--grammar"}, "GRAMMAR", - format("BNF-like grammar to constrain generations (see samples in grammars/ dir) (default: '%s')", params.sparams.grammar.c_str()), + string_format("BNF-like grammar to constrain generations (see samples in grammars/ dir) (default: '%s')", params.sparams.grammar.c_str()), [](common_params & params, const std::string & value) { params.sparams.grammar = value; } @@ -1080,7 +1054,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex [](common_params & params, const std::string & value) { std::ifstream file(value); if (!file) { - throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str())); + throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str())); } std::copy( std::istreambuf_iterator(file), @@ -1150,49 +1124,49 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_env("LLAMA_ARG_ROPE_FREQ_SCALE")); add_opt(common_arg( {"--yarn-orig-ctx"}, "N", - format("YaRN: original context size of model (default: %d = model training context size)", params.yarn_orig_ctx), + string_format("YaRN: original context size of model (default: %d = model training context size)", params.yarn_orig_ctx), [](common_params & params, int value) { params.yarn_orig_ctx = value; } ).set_env("LLAMA_ARG_YARN_ORIG_CTX")); add_opt(common_arg( {"--yarn-ext-factor"}, "N", - format("YaRN: extrapolation mix factor (default: %.1f, 0.0 = full interpolation)", (double)params.yarn_ext_factor), + string_format("YaRN: extrapolation mix factor (default: %.1f, 0.0 = full interpolation)", (double)params.yarn_ext_factor), [](common_params & params, const std::string & value) { params.yarn_ext_factor = std::stof(value); } ).set_env("LLAMA_ARG_YARN_EXT_FACTOR")); add_opt(common_arg( {"--yarn-attn-factor"}, "N", - format("YaRN: scale sqrt(t) or attention magnitude (default: %.1f)", (double)params.yarn_attn_factor), + string_format("YaRN: scale sqrt(t) or attention magnitude (default: %.1f)", (double)params.yarn_attn_factor), [](common_params & params, const std::string & value) { params.yarn_attn_factor = std::stof(value); } ).set_env("LLAMA_ARG_YARN_ATTN_FACTOR")); add_opt(common_arg( {"--yarn-beta-slow"}, "N", - format("YaRN: high correction dim or alpha (default: %.1f)", (double)params.yarn_beta_slow), + string_format("YaRN: high correction dim or alpha (default: %.1f)", (double)params.yarn_beta_slow), [](common_params & params, const std::string & value) { params.yarn_beta_slow = std::stof(value); } ).set_env("LLAMA_ARG_YARN_BETA_SLOW")); add_opt(common_arg( {"--yarn-beta-fast"}, "N", - format("YaRN: low correction dim or beta (default: %.1f)", (double)params.yarn_beta_fast), + string_format("YaRN: low correction dim or beta (default: %.1f)", (double)params.yarn_beta_fast), [](common_params & params, const std::string & value) { params.yarn_beta_fast = std::stof(value); } ).set_env("LLAMA_ARG_YARN_BETA_FAST")); add_opt(common_arg( {"-gan", "--grp-attn-n"}, "N", - format("group-attention factor (default: %d)", params.grp_attn_n), + string_format("group-attention factor (default: %d)", params.grp_attn_n), [](common_params & params, int value) { params.grp_attn_n = value; } ).set_env("LLAMA_ARG_GRP_ATTN_N")); add_opt(common_arg( {"-gaw", "--grp-attn-w"}, "N", - format("group-attention width (default: %.1f)", (double)params.grp_attn_w), + string_format("group-attention width (default: %.1f)", (double)params.grp_attn_w), [](common_params & params, int value) { params.grp_attn_w = value; } @@ -1213,7 +1187,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_env("LLAMA_ARG_NO_KV_OFFLOAD")); add_opt(common_arg( {"-ctk", "--cache-type-k"}, "TYPE", - format("KV cache data type for K (default: %s)", params.cache_type_k.c_str()), + string_format("KV cache data type for K (default: %s)", params.cache_type_k.c_str()), [](common_params & params, const std::string & value) { // TODO: get the type right here params.cache_type_k = value; @@ -1221,7 +1195,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_env("LLAMA_ARG_CACHE_TYPE_K")); add_opt(common_arg( {"-ctv", "--cache-type-v"}, "TYPE", - format("KV cache data type for V (default: %s)", params.cache_type_v.c_str()), + string_format("KV cache data type for V (default: %s)", params.cache_type_v.c_str()), [](common_params & params, const std::string & value) { // TODO: get the type right here params.cache_type_v = value; @@ -1229,7 +1203,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_env("LLAMA_ARG_CACHE_TYPE_V")); add_opt(common_arg( {"--perplexity", "--all-logits"}, - format("return logits for all tokens in the batch (default: %s)", params.logits_all ? "true" : "false"), + string_format("return logits for all tokens in the batch (default: %s)", params.logits_all ? "true" : "false"), [](common_params & params) { params.logits_all = true; } @@ -1243,7 +1217,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_PERPLEXITY})); add_opt(common_arg( {"--hellaswag-tasks"}, "N", - format("number of tasks to use when computing the HellaSwag score (default: %zu)", params.hellaswag_tasks), + string_format("number of tasks to use when computing the HellaSwag score (default: %zu)", params.hellaswag_tasks), [](common_params & params, int value) { params.hellaswag_tasks = value; } @@ -1257,7 +1231,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_PERPLEXITY})); add_opt(common_arg( {"--winogrande-tasks"}, "N", - format("number of tasks to use when computing the Winogrande score (default: %zu)", params.winogrande_tasks), + string_format("number of tasks to use when computing the Winogrande score (default: %zu)", params.winogrande_tasks), [](common_params & params, int value) { params.winogrande_tasks = value; } @@ -1271,7 +1245,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_PERPLEXITY})); add_opt(common_arg( {"--multiple-choice-tasks"}, "N", - format("number of tasks to use when computing the multiple choice score (default: %zu)", params.multiple_choice_tasks), + string_format("number of tasks to use when computing the multiple choice score (default: %zu)", params.multiple_choice_tasks), [](common_params & params, int value) { params.multiple_choice_tasks = value; } @@ -1292,42 +1266,42 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_PERPLEXITY})); add_opt(common_arg( {"--ppl-stride"}, "N", - format("stride for perplexity calculation (default: %d)", params.ppl_stride), + string_format("stride for perplexity calculation (default: %d)", params.ppl_stride), [](common_params & params, int value) { params.ppl_stride = value; } ).set_examples({LLAMA_EXAMPLE_PERPLEXITY})); add_opt(common_arg( {"--ppl-output-type"}, "<0|1>", - format("output type for perplexity calculation (default: %d)", params.ppl_output_type), + string_format("output type for perplexity calculation (default: %d)", params.ppl_output_type), [](common_params & params, int value) { params.ppl_output_type = value; } ).set_examples({LLAMA_EXAMPLE_PERPLEXITY})); add_opt(common_arg( {"-dt", "--defrag-thold"}, "N", - format("KV cache defragmentation threshold (default: %.1f, < 0 - disabled)", (double)params.defrag_thold), + string_format("KV cache defragmentation threshold (default: %.1f, < 0 - disabled)", (double)params.defrag_thold), [](common_params & params, const std::string & value) { params.defrag_thold = std::stof(value); } ).set_env("LLAMA_ARG_DEFRAG_THOLD")); add_opt(common_arg( {"-np", "--parallel"}, "N", - format("number of parallel sequences to decode (default: %d)", params.n_parallel), + string_format("number of parallel sequences to decode (default: %d)", params.n_parallel), [](common_params & params, int value) { params.n_parallel = value; } ).set_env("LLAMA_ARG_N_PARALLEL")); add_opt(common_arg( {"-ns", "--sequences"}, "N", - format("number of sequences to decode (default: %d)", params.n_sequences), + string_format("number of sequences to decode (default: %d)", params.n_sequences), [](common_params & params, int value) { params.n_sequences = value; } ).set_examples({LLAMA_EXAMPLE_PARALLEL})); add_opt(common_arg( {"-cb", "--cont-batching"}, - format("enable continuous batching (a.k.a dynamic batching) (default: %s)", params.cont_batching ? "enabled" : "disabled"), + string_format("enable continuous batching (a.k.a dynamic batching) (default: %s)", params.cont_batching ? "enabled" : "disabled"), [](common_params & params) { params.cont_batching = true; } @@ -1451,7 +1425,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex std::vector split_arg{ it, {} }; if (split_arg.size() >= llama_max_devices()) { throw std::invalid_argument( - format("got %d input configs, but system only has %d devices", (int)split_arg.size(), (int)llama_max_devices()) + string_format("got %d input configs, but system only has %d devices", (int)split_arg.size(), (int)llama_max_devices()) ); } for (size_t i = 0; i < llama_max_devices(); ++i) { @@ -1468,7 +1442,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_env("LLAMA_ARG_TENSOR_SPLIT")); add_opt(common_arg( {"-mg", "--main-gpu"}, "INDEX", - format("the GPU to use for the model (with split-mode = none), or for intermediate results and KV (with split-mode = row) (default: %d)", params.main_gpu), + string_format("the GPU to use for the model (with split-mode = none), or for intermediate results and KV (with split-mode = row) (default: %d)", params.main_gpu), [](common_params & params, int value) { params.main_gpu = value; if (!llama_supports_gpu_offload()) { @@ -1478,7 +1452,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_env("LLAMA_ARG_MAIN_GPU")); add_opt(common_arg( {"--check-tensors"}, - format("check model tensor data for invalid values (default: %s)", params.check_tensors ? "true" : "false"), + string_format("check model tensor data for invalid values (default: %s)", params.check_tensors ? "true" : "false"), [](common_params & params) { params.check_tensors = true; } @@ -1489,7 +1463,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex "types: int, float, bool, str. example: --override-kv tokenizer.ggml.add_bos_token=bool:false", [](common_params & params, const std::string & value) { if (!string_parse_kv_override(value.c_str(), params.kv_overrides)) { - throw std::runtime_error(format("error: Invalid type for KV override: %s\n", value.c_str())); + throw std::runtime_error(string_format("error: Invalid type for KV override: %s\n", value.c_str())); } } )); @@ -1543,7 +1517,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex {"-m", "--model"}, "FNAME", ex == LLAMA_EXAMPLE_EXPORT_LORA ? std::string("model path from which to load base model") - : format( + : string_format( "model path (default: `models/$filename` with filename from `--hf-file` " "or `--model-url` if set, otherwise %s)", DEFAULT_MODEL_PATH ), @@ -1592,42 +1566,42 @@ common_params_context common_params_parser_init(common_params & params, llama_ex [](common_params & params, const std::string & value) { std::ifstream file(value, std::ios::binary); if (!file) { - throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str())); + throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str())); } params.context_files.push_back(value); } ).set_examples({LLAMA_EXAMPLE_RETRIEVAL})); add_opt(common_arg( {"--chunk-size"}, "N", - format("minimum length of embedded text chunks (default: %d)", params.chunk_size), + string_format("minimum length of embedded text chunks (default: %d)", params.chunk_size), [](common_params & params, int value) { params.chunk_size = value; } ).set_examples({LLAMA_EXAMPLE_RETRIEVAL})); add_opt(common_arg( {"--chunk-separator"}, "STRING", - format("separator between chunks (default: '%s')", params.chunk_separator.c_str()), + string_format("separator between chunks (default: '%s')", params.chunk_separator.c_str()), [](common_params & params, const std::string & value) { params.chunk_separator = value; } ).set_examples({LLAMA_EXAMPLE_RETRIEVAL})); add_opt(common_arg( {"--junk"}, "N", - format("number of times to repeat the junk text (default: %d)", params.n_junk), + string_format("number of times to repeat the junk text (default: %d)", params.n_junk), [](common_params & params, int value) { params.n_junk = value; } ).set_examples({LLAMA_EXAMPLE_PASSKEY})); add_opt(common_arg( {"--pos"}, "N", - format("position of the passkey in the junk text (default: %d)", params.i_pos), + string_format("position of the passkey in the junk text (default: %d)", params.i_pos), [](common_params & params, int value) { params.i_pos = value; } ).set_examples({LLAMA_EXAMPLE_PASSKEY})); add_opt(common_arg( {"-o", "--output", "--output-file"}, "FNAME", - format("output file (default: '%s')", + string_format("output file (default: '%s')", ex == LLAMA_EXAMPLE_EXPORT_LORA ? params.lora_outfile.c_str() : ex == LLAMA_EXAMPLE_CVECTOR_GENERATOR @@ -1641,42 +1615,42 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_CVECTOR_GENERATOR, LLAMA_EXAMPLE_EXPORT_LORA})); add_opt(common_arg( {"-ofreq", "--output-frequency"}, "N", - format("output the imatrix every N iterations (default: %d)", params.n_out_freq), + string_format("output the imatrix every N iterations (default: %d)", params.n_out_freq), [](common_params & params, int value) { params.n_out_freq = value; } ).set_examples({LLAMA_EXAMPLE_IMATRIX})); add_opt(common_arg( {"--save-frequency"}, "N", - format("save an imatrix copy every N iterations (default: %d)", params.n_save_freq), + string_format("save an imatrix copy every N iterations (default: %d)", params.n_save_freq), [](common_params & params, int value) { params.n_save_freq = value; } ).set_examples({LLAMA_EXAMPLE_IMATRIX})); add_opt(common_arg( {"--process-output"}, - format("collect data for the output tensor (default: %s)", params.process_output ? "true" : "false"), + string_format("collect data for the output tensor (default: %s)", params.process_output ? "true" : "false"), [](common_params & params) { params.process_output = true; } ).set_examples({LLAMA_EXAMPLE_IMATRIX})); add_opt(common_arg( {"--no-ppl"}, - format("do not compute perplexity (default: %s)", params.compute_ppl ? "true" : "false"), + string_format("do not compute perplexity (default: %s)", params.compute_ppl ? "true" : "false"), [](common_params & params) { params.compute_ppl = false; } ).set_examples({LLAMA_EXAMPLE_IMATRIX})); add_opt(common_arg( {"--chunk", "--from-chunk"}, "N", - format("start processing the input from chunk N (default: %d)", params.i_chunk), + string_format("start processing the input from chunk N (default: %d)", params.i_chunk), [](common_params & params, int value) { params.i_chunk = value; } ).set_examples({LLAMA_EXAMPLE_IMATRIX})); add_opt(common_arg( {"-pps"}, - format("is the prompt shared across parallel sequences (default: %s)", params.is_pp_shared ? "true" : "false"), + string_format("is the prompt shared across parallel sequences (default: %s)", params.is_pp_shared ? "true" : "false"), [](common_params & params) { params.is_pp_shared = true; } @@ -1707,7 +1681,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_BENCH})); add_opt(common_arg( {"--embd-normalize"}, "N", - format("normalisation for embendings (default: %d) (-1=none, 0=max absolute int16, 1=taxicab, 2=euclidean, >2=p-norm)", params.embd_normalize), + string_format("normalisation for embendings (default: %d) (-1=none, 0=max absolute int16, 1=taxicab, 2=euclidean, >2=p-norm)", params.embd_normalize), [](common_params & params, int value) { params.embd_normalize = value; } @@ -1728,35 +1702,35 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_EMBEDDING})); add_opt(common_arg( {"--host"}, "HOST", - format("ip address to listen (default: %s)", params.hostname.c_str()), + string_format("ip address to listen (default: %s)", params.hostname.c_str()), [](common_params & params, const std::string & value) { params.hostname = value; } ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_HOST")); add_opt(common_arg( {"--port"}, "PORT", - format("port to listen (default: %d)", params.port), + string_format("port to listen (default: %d)", params.port), [](common_params & params, int value) { params.port = value; } ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_PORT")); add_opt(common_arg( {"--path"}, "PATH", - format("path to serve static files from (default: %s)", params.public_path.c_str()), + string_format("path to serve static files from (default: %s)", params.public_path.c_str()), [](common_params & params, const std::string & value) { params.public_path = value; } ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_STATIC_PATH")); add_opt(common_arg( {"--embedding", "--embeddings"}, - format("restrict to only support embedding use case; use only with dedicated embedding models (default: %s)", params.embedding ? "enabled" : "disabled"), + string_format("restrict to only support embedding use case; use only with dedicated embedding models (default: %s)", params.embedding ? "enabled" : "disabled"), [](common_params & params) { params.embedding = true; } ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_EMBEDDINGS")); add_opt(common_arg( {"--reranking", "--rerank"}, - format("enable reranking endpoint on server (default: %s)", params.reranking ? "enabled" : "disabled"), + string_format("enable reranking endpoint on server (default: %s)", params.reranking ? "enabled" : "disabled"), [](common_params & params) { params.reranking = true; } @@ -1774,7 +1748,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex [](common_params & params, const std::string & value) { std::ifstream key_file(value); if (!key_file) { - throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str())); + throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str())); } std::string key; while (std::getline(key_file, key)) { @@ -1801,7 +1775,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_SSL_CERT_FILE")); add_opt(common_arg( {"-to", "--timeout"}, "N", - format("server read/write timeout in seconds (default: %d)", params.timeout_read), + string_format("server read/write timeout in seconds (default: %d)", params.timeout_read), [](common_params & params, int value) { params.timeout_read = value; params.timeout_write = value; @@ -1809,7 +1783,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_TIMEOUT")); add_opt(common_arg( {"--threads-http"}, "N", - format("number of threads used to process HTTP requests (default: %d)", params.n_threads_http), + string_format("number of threads used to process HTTP requests (default: %d)", params.n_threads_http), [](common_params & params, int value) { params.n_threads_http = value; } @@ -1820,7 +1794,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex [](common_params & params, const std::string & value) { std::ifstream file(value); if (!file) { - throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str())); + throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str())); } std::string system_prompt; std::copy( @@ -1833,21 +1807,21 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_SERVER})); add_opt(common_arg( {"--metrics"}, - format("enable prometheus compatible metrics endpoint (default: %s)", params.endpoint_metrics ? "enabled" : "disabled"), + string_format("enable prometheus compatible metrics endpoint (default: %s)", params.endpoint_metrics ? "enabled" : "disabled"), [](common_params & params) { params.endpoint_metrics = true; } ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_METRICS")); add_opt(common_arg( {"--slots"}, - format("enable slots monitoring endpoint (default: %s)", params.endpoint_slots ? "enabled" : "disabled"), + string_format("enable slots monitoring endpoint (default: %s)", params.endpoint_slots ? "enabled" : "disabled"), [](common_params & params) { params.endpoint_slots = true; } ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_SLOTS")); add_opt(common_arg( {"--props"}, - format("enable changing global properties via POST /props (default: %s)", params.endpoint_props ? "enabled" : "disabled"), + string_format("enable changing global properties via POST /props (default: %s)", params.endpoint_props ? "enabled" : "disabled"), [](common_params & params) { params.endpoint_props = true; } @@ -1877,7 +1851,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex "only commonly used templates are accepted:\nhttps://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template", [](common_params & params, const std::string & value) { if (!common_chat_verify_template(value)) { - throw std::runtime_error(format( + throw std::runtime_error(string_format( "error: the supplied chat template is not supported: %s\n" "note: llama.cpp does not use jinja parser, we only support commonly used templates\n", value.c_str() @@ -1888,14 +1862,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CHAT_TEMPLATE")); add_opt(common_arg( {"-sps", "--slot-prompt-similarity"}, "SIMILARITY", - format("how much the prompt of a request must match the prompt of a slot in order to use that slot (default: %.2f, 0.0 = disabled)\n", params.slot_prompt_similarity), + string_format("how much the prompt of a request must match the prompt of a slot in order to use that slot (default: %.2f, 0.0 = disabled)\n", params.slot_prompt_similarity), [](common_params & params, const std::string & value) { params.slot_prompt_similarity = std::stof(value); } ).set_examples({LLAMA_EXAMPLE_SERVER})); add_opt(common_arg( {"--lora-init-without-apply"}, - format("load LoRA adapters without applying them (apply later via POST /lora-adapters) (default: %s)", params.lora_init_without_apply ? "enabled" : "disabled"), + string_format("load LoRA adapters without applying them (apply later via POST /lora-adapters) (default: %s)", params.lora_init_without_apply ? "enabled" : "disabled"), [](common_params & params) { params.lora_init_without_apply = true; } @@ -1920,28 +1894,28 @@ common_params_context common_params_parser_init(common_params & params, llama_ex )); add_opt(common_arg( {"--positive-file"}, "FNAME", - format("positive prompts file, one prompt per line (default: '%s')", params.cvector_positive_file.c_str()), + string_format("positive prompts file, one prompt per line (default: '%s')", params.cvector_positive_file.c_str()), [](common_params & params, const std::string & value) { params.cvector_positive_file = value; } ).set_examples({LLAMA_EXAMPLE_CVECTOR_GENERATOR})); add_opt(common_arg( {"--negative-file"}, "FNAME", - format("negative prompts file, one prompt per line (default: '%s')", params.cvector_negative_file.c_str()), + string_format("negative prompts file, one prompt per line (default: '%s')", params.cvector_negative_file.c_str()), [](common_params & params, const std::string & value) { params.cvector_negative_file = value; } ).set_examples({LLAMA_EXAMPLE_CVECTOR_GENERATOR})); add_opt(common_arg( {"--pca-batch"}, "N", - format("batch size used for PCA. Larger batch runs faster, but uses more memory (default: %d)", params.n_pca_batch), + string_format("batch size used for PCA. Larger batch runs faster, but uses more memory (default: %d)", params.n_pca_batch), [](common_params & params, int value) { params.n_pca_batch = value; } ).set_examples({LLAMA_EXAMPLE_CVECTOR_GENERATOR})); add_opt(common_arg( {"--pca-iter"}, "N", - format("number of iterations used for PCA (default: %d)", params.n_pca_iterations), + string_format("number of iterations used for PCA (default: %d)", params.n_pca_iterations), [](common_params & params, int value) { params.n_pca_iterations = value; } diff --git a/common/common.cpp b/common/common.cpp index d1b92250a..451307b55 100644 --- a/common/common.cpp +++ b/common/common.cpp @@ -12,6 +12,7 @@ #include #include +#include #include #include #include @@ -23,10 +24,10 @@ #include #include #include +#include #include #include #include -#include #if defined(__APPLE__) && defined(__MACH__) #include @@ -400,6 +401,21 @@ std::string common_params_get_system_info(const common_params & params) { // String utils // +std::string string_format(const char * fmt, ...) { + va_list ap; + va_list ap2; + va_start(ap, fmt); + va_copy(ap2, ap); + int size = vsnprintf(NULL, 0, fmt, ap); + GGML_ASSERT(size >= 0 && size < INT_MAX); // NOLINT + std::vector buf(size + 1); + int size2 = vsnprintf(buf.data(), size + 1, fmt, ap2); + GGML_ASSERT(size2 == size); + va_end(ap2); + va_end(ap); + return std::string(buf.data(), size); +} + std::vector string_split(std::string input, char separator) { std::vector parts; size_t separator_pos = input.find(separator); diff --git a/common/common.h b/common/common.h index ea2719e4b..5beec4bde 100644 --- a/common/common.h +++ b/common/common.h @@ -352,15 +352,28 @@ void common_init(); std::string common_params_get_system_info(const common_params & params); -bool parse_cpu_range(const std::string& range, bool(&boolmask)[GGML_MAX_N_THREADS]); -bool parse_cpu_mask(const std::string& mask, bool(&boolmask)[GGML_MAX_N_THREADS]); -void postprocess_cpu_params(cpu_params& cpuparams, const cpu_params* role_model = nullptr); +bool parse_cpu_range(const std::string & range, bool(&boolmask)[GGML_MAX_N_THREADS]); +bool parse_cpu_mask(const std::string & mask, bool(&boolmask)[GGML_MAX_N_THREADS]); +void postprocess_cpu_params(cpu_params & cpuparams, const cpu_params * role_model = nullptr); bool set_process_priority(enum ggml_sched_priority prio); // // String utils // +#ifdef __GNUC__ +#ifdef __MINGW32__ +#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(gnu_printf, __VA_ARGS__))) +#else +#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__))) +#endif +#else +#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) +#endif + +LLAMA_COMMON_ATTRIBUTE_FORMAT(1, 2) +std::string string_format(const char * fmt, ...); + std::vector string_split(std::string input, char separator); std::string string_strip(const std::string & str); diff --git a/examples/infill/infill.cpp b/examples/infill/infill.cpp index 3d0f71fda..f82c614f5 100644 --- a/examples/infill/infill.cpp +++ b/examples/infill/infill.cpp @@ -205,11 +205,11 @@ int main(int argc, char ** argv) { std::vector inp_pfx = common_tokenize(ctx, params.input_prefix, false); std::vector inp_sfx = common_tokenize(ctx, params.input_suffix, false); - GGML_ASSERT(llama_token_prefix(model) >= 0); - GGML_ASSERT(llama_token_suffix(model) >= 0); + GGML_ASSERT(llama_token_fim_pre(model) >= 0); + GGML_ASSERT(llama_token_fim_suf(model) >= 0); - inp_pfx.insert(inp_pfx.begin(), llama_token_prefix(model)); - inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model)); + inp_pfx.insert(inp_pfx.begin(), llama_token_fim_pre(model)); + inp_sfx.insert(inp_sfx.begin(), llama_token_fim_suf(model)); embd_inp = params.spm_infill ? inp_sfx : inp_pfx; embd_end = params.spm_infill ? inp_pfx : inp_sfx; @@ -218,7 +218,7 @@ int main(int argc, char ** argv) { } embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end()); - const llama_token middle_token = llama_token_middle(model); + const llama_token middle_token = llama_token_fim_mid(model); if (middle_token >= 0) { embd_inp.push_back(middle_token); } @@ -508,8 +508,8 @@ int main(int argc, char ** argv) { std::vector inp_pfx = common_tokenize(ctx, params.input_prefix, false); std::vector inp_sfx = common_tokenize(ctx, params.input_suffix, false); - inp_pfx.insert(inp_pfx.begin(), llama_token_prefix(model)); - inp_sfx.insert(inp_sfx.begin(), llama_token_suffix(model)); + inp_pfx.insert(inp_pfx.begin(), llama_token_fim_pre(model)); + inp_sfx.insert(inp_sfx.begin(), llama_token_fim_suf(model)); embd_inp = params.spm_infill ? inp_sfx : inp_pfx; embd_end = params.spm_infill ? inp_pfx : inp_sfx; diff --git a/examples/server/README.md b/examples/server/README.md index 09d1cf097..3da0130ac 100644 --- a/examples/server/README.md +++ b/examples/server/README.md @@ -526,7 +526,7 @@ Takes a prefix and a suffix and returns the predicted completion as stream. - `input_prefix`: Set the prefix of the code to infill. - `input_suffix`: Set the suffix of the code to infill. -It also accepts all the options of `/completion` except `stream` and `prompt`. +It also accepts all the options of `/completion`. ### **GET** `/props`: Get server global properties. diff --git a/examples/server/server.cpp b/examples/server/server.cpp index 2e1d24189..314a506a1 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -753,12 +753,7 @@ struct server_context { metrics.init(); } - std::vector tokenize(const json & json_prompt, bool add_special) const { - // TODO: currently, we tokenize using special tokens by default - // this is not always correct (see https://github.com/ggerganov/llama.cpp/pull/4160#issuecomment-1824826216) - // but it's better compared to completely ignoring ChatML and other chat templates - const bool TMP_FORCE_SPECIAL = true; - + std::vector tokenize(const json & json_prompt, bool add_special, bool parse_special) const { // If `add_bos` is true, we only add BOS, when json_prompt is a string, // or the first element of the json_prompt array is a string. std::vector prompt_tokens; @@ -771,10 +766,10 @@ struct server_context { std::vector p; if (first) { - p = common_tokenize(ctx, s, add_special, TMP_FORCE_SPECIAL); + p = common_tokenize(ctx, s, add_special, parse_special); first = false; } else { - p = common_tokenize(ctx, s, false, TMP_FORCE_SPECIAL); + p = common_tokenize(ctx, s, false, parse_special); } prompt_tokens.insert(prompt_tokens.end(), p.begin(), p.end()); @@ -788,7 +783,7 @@ struct server_context { } } else { auto s = json_prompt.template get(); - prompt_tokens = common_tokenize(ctx, s, add_special, TMP_FORCE_SPECIAL); + prompt_tokens = common_tokenize(ctx, s, add_special, parse_special); } return prompt_tokens; @@ -1215,7 +1210,7 @@ struct server_context { slot.params.n_predict, n_ctx_train); } - SLT_DBG(slot, "n_decoded = %d, n_remaining = %d, next token: '%s'\n", slot.n_decoded, slot.n_remaining, token_str.c_str()); + SLT_DBG(slot, "n_decoded = %d, n_remaining = %d, next token: %5d '%s'\n", slot.n_decoded, slot.n_remaining, result.tok, token_str.c_str()); return slot.has_next_token; // continue } @@ -1483,9 +1478,8 @@ struct server_context { if (prompt.is_string() || json_is_array_of_numbers(prompt)) { data["index"] = 0; create_task(data, false, nullptr); - } - // otherwise, it's a multiple-prompt task, we break it into smaller tasks - else if (prompt.is_array()) { + } else if (prompt.is_array()) { + // otherwise, it's a multiple-prompt task, we break it into smaller tasks std::vector prompts = prompt; if (cmpl_type == SERVER_TASK_CMPL_TYPE_RERANK) { // prompts[0] is the question @@ -1510,9 +1504,8 @@ struct server_context { } } } - } - // invalid case - else { + } else { + // invalid case throw std::runtime_error(error_msg); } @@ -1785,6 +1778,9 @@ struct server_context { } slot->cache_tokens.resize(token_count); + // TODO: maybe detokenize the slot->cache_tokens instead? + slot->prompt = string_format("[restored %d tokens from file]", (int) token_count); + const int64_t t_end = ggml_time_us(); const double t_restore_ms = (t_end - t_start) / 1000.0; @@ -1971,63 +1967,57 @@ struct server_context { slot.t_start_process_prompt = ggml_time_us(); slot.t_start_generation = 0; - if (slot.cmpl_type == SERVER_TASK_CMPL_TYPE_INFILL) { - const bool add_bos = llama_add_bos_token(model); - bool suff_rm_leading_spc = true; - if (params.input_suffix.find_first_of(' ') == 0 && params.input_suffix.size() > 1) { - params.input_suffix.erase(0, 1); - suff_rm_leading_spc = false; - } + switch (slot.cmpl_type) { + case SERVER_TASK_CMPL_TYPE_NORMAL: + case SERVER_TASK_CMPL_TYPE_EMBEDDING: + { + prompt_tokens = tokenize(slot.prompt, system_prompt.empty(), true); // add BOS if there isn't system prompt + } break; + case SERVER_TASK_CMPL_TYPE_RERANK: + { + // require slot.prompt to be array of 2 strings + if (!slot.prompt.is_array() || slot.prompt.size() != 2) { + SLT_ERR(slot, "%s", "invalid prompt for rerank task\n"); + slot.release(); + send_error(slot, "invalid prompt for rerank task", ERROR_TYPE_INVALID_REQUEST); + continue; + } - auto prefix_tokens = tokenize(slot.params.input_prefix, false); - auto suffix_tokens = tokenize(slot.params.input_suffix, false); + // prompt: [BOS]query[EOS][SEP]doc[EOS] + prompt_tokens.clear(); + prompt_tokens.push_back(llama_token_bos(model)); + { + const auto part = tokenize(slot.prompt[0], false, false); + prompt_tokens.insert(prompt_tokens.end(), part.begin(), part.end()); + } + prompt_tokens.push_back(llama_token_eos(model)); + prompt_tokens.push_back(llama_token_sep(model)); + { + const auto part = tokenize(slot.prompt[1], false, false); + prompt_tokens.insert(prompt_tokens.end(), part.begin(), part.end()); + } + prompt_tokens.push_back(llama_token_eos(model)); + } break; + case SERVER_TASK_CMPL_TYPE_INFILL: + { + auto prefix_tokens = tokenize(slot.params.input_prefix, false, false); + auto suffix_tokens = tokenize(slot.params.input_suffix, false, false); - const int space_token = 29871; // TODO: this should not be hardcoded - if (suff_rm_leading_spc && !suffix_tokens.empty() && suffix_tokens[0] == space_token) { - suffix_tokens.erase(suffix_tokens.begin()); - } + prefix_tokens.insert(prefix_tokens.begin(), llama_token_fim_pre(model)); + suffix_tokens.insert(suffix_tokens.begin(), llama_token_fim_suf(model)); - prefix_tokens.insert(prefix_tokens.begin(), llama_token_prefix(model)); - suffix_tokens.insert(suffix_tokens.begin(), llama_token_suffix(model)); + auto embd_inp = params.spm_infill ? suffix_tokens : prefix_tokens; + auto embd_end = params.spm_infill ? prefix_tokens : suffix_tokens; - auto embd_inp = params.spm_infill ? suffix_tokens : prefix_tokens; - auto embd_end = params.spm_infill ? prefix_tokens : suffix_tokens; - if (add_bos) { - embd_inp.insert(embd_inp.begin(), llama_token_bos(model)); - } - embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end()); + if (llama_add_bos_token(model)) { + embd_inp.insert(embd_inp.begin(), llama_token_bos(model)); + } - const llama_token middle_token = llama_token_middle(model); - if (middle_token >= 0) { - embd_inp.push_back(middle_token); - } + embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end()); + embd_inp.push_back(llama_token_fim_mid(model)); - prompt_tokens = embd_inp; - } else if (slot.cmpl_type == SERVER_TASK_CMPL_TYPE_RERANK) { - // require slot.prompt to be array of 2 strings - if (!slot.prompt.is_array() || slot.prompt.size() != 2) { - SLT_ERR(slot, "%s", "invalid prompt for rerank task\n"); - slot.release(); - send_error(slot, "invalid prompt for rerank task", ERROR_TYPE_INVALID_REQUEST); - continue; - } - - // prompt: [BOS]query[EOS][SEP]doc[EOS] - prompt_tokens.clear(); - prompt_tokens.push_back(llama_token_bos(model)); - { - const auto part = tokenize(slot.prompt[0], false); - prompt_tokens.insert(prompt_tokens.end(), part.begin(), part.end()); - } - prompt_tokens.push_back(llama_token_eos(model)); - prompt_tokens.push_back(llama_token_sep(model)); - { - const auto part = tokenize(slot.prompt[1], false); - prompt_tokens.insert(prompt_tokens.end(), part.begin(), part.end()); - } - prompt_tokens.push_back(llama_token_eos(model)); - } else { - prompt_tokens = tokenize(slot.prompt, system_prompt.empty()); // add BOS if there isn't system prompt + prompt_tokens = std::move(embd_inp); + } break; } slot.n_past = 0; @@ -2035,6 +2025,11 @@ struct server_context { SLT_INF(slot, "prompt tokenized, n_ctx_slot = %d, n_keep = %d, n_prompt_tokens = %d\n", slot.n_ctx, slot.params.n_keep, slot.n_prompt_tokens); + // print prompt tokens: + for (int i = 0; i < (int) prompt_tokens.size(); i++) { + SLT_DBG(slot, "prompt token %3d: %6d '%s'\n", i, prompt_tokens[i], common_token_to_piece(ctx, prompt_tokens[i]).c_str()); + } + // empty prompt passed -> release the slot and send empty response if (prompt_tokens.empty()) { SLT_WRN(slot, "%s", "empty prompt - releasing slot\n"); @@ -2924,7 +2919,23 @@ int main(int argc, char ** argv) { return handle_completions_generic(SERVER_TASK_CMPL_TYPE_NORMAL, data, res); }; - const auto handle_infill = [&handle_completions_generic](const httplib::Request & req, httplib::Response & res) { + const auto handle_infill = [&ctx_server, &res_error, &handle_completions_generic](const httplib::Request & req, httplib::Response & res) { + std::string err; + if (llama_token_fim_pre(ctx_server.model) == LLAMA_TOKEN_NULL) { + err += "prefix token is missing. "; + } + if (llama_token_fim_suf(ctx_server.model) == LLAMA_TOKEN_NULL) { + err += "suffix token is missing. "; + } + if (llama_token_fim_mid(ctx_server.model) == LLAMA_TOKEN_NULL) { + err += "middle token is missing. "; + } + + if (!err.empty()) { + res_error(res, format_error_response(string_format("Infill is not supported by this model: %s", err.c_str()), ERROR_TYPE_NOT_SUPPORTED)); + return; + } + json data = json::parse(req.body); return handle_completions_generic(SERVER_TASK_CMPL_TYPE_INFILL, data, res); }; @@ -3010,7 +3021,8 @@ int main(int argc, char ** argv) { if (body.count("content") != 0) { const bool add_special = json_value(body, "add_special", false); const bool with_pieces = json_value(body, "with_pieces", false); - std::vector tokens = ctx_server.tokenize(body.at("content"), add_special); + + std::vector tokens = ctx_server.tokenize(body.at("content"), add_special, true); if (with_pieces) { for (const auto& token : tokens) { diff --git a/gguf-py/gguf/constants.py b/gguf-py/gguf/constants.py index e08617ba2..7ab08b036 100644 --- a/gguf-py/gguf/constants.py +++ b/gguf-py/gguf/constants.py @@ -152,6 +152,8 @@ class Keys: MERGES = "tokenizer.ggml.merges" BOS_ID = "tokenizer.ggml.bos_token_id" EOS_ID = "tokenizer.ggml.eos_token_id" + EOT_ID = "tokenizer.ggml.eot_token_id" + EOM_ID = "tokenizer.ggml.eom_token_id" UNK_ID = "tokenizer.ggml.unknown_token_id" SEP_ID = "tokenizer.ggml.seperator_token_id" PAD_ID = "tokenizer.ggml.padding_token_id" @@ -168,11 +170,16 @@ class Keys: CHAT_TEMPLATE_N = "tokenizer.chat_template.{name}" CHAT_TEMPLATES = "tokenizer.chat_templates" # FIM/Infill special tokens constants + FIM_PRE_ID = "tokenizer.ggml.fim_pre_token_id" + FIM_SUF_ID = "tokenizer.ggml.fim_suf_token_id" + FIM_MID_ID = "tokenizer.ggml.fim_mid_token_id" + FIM_PAD_ID = "tokenizer.ggml.fim_pad_token_id" + FIM_REP_ID = "tokenizer.ggml.fim_rep_token_id" + FIM_SEP_ID = "tokenizer.ggml.fim_sep_token_id" + # deprecated: PREFIX_ID = "tokenizer.ggml.prefix_token_id" SUFFIX_ID = "tokenizer.ggml.suffix_token_id" MIDDLE_ID = "tokenizer.ggml.middle_token_id" - EOT_ID = "tokenizer.ggml.eot_token_id" - EOM_ID = "tokenizer.ggml.eom_token_id" class Adapter: TYPE = "adapter.type" @@ -1579,6 +1586,8 @@ KEY_TOKENIZER_SCORES = Keys.Tokenizer.SCORES KEY_TOKENIZER_MERGES = Keys.Tokenizer.MERGES KEY_TOKENIZER_BOS_ID = Keys.Tokenizer.BOS_ID KEY_TOKENIZER_EOS_ID = Keys.Tokenizer.EOS_ID +KEY_TOKENIZER_EOT_ID = Keys.Tokenizer.EOT_ID +KEY_TOKENIZER_EOM_ID = Keys.Tokenizer.EOM_ID KEY_TOKENIZER_UNK_ID = Keys.Tokenizer.UNK_ID KEY_TOKENIZER_SEP_ID = Keys.Tokenizer.SEP_ID KEY_TOKENIZER_PAD_ID = Keys.Tokenizer.PAD_ID @@ -1586,8 +1595,15 @@ KEY_TOKENIZER_CLS_ID = Keys.Tokenizer.CLS_ID KEY_TOKENIZER_MASK_ID = Keys.Tokenizer.MASK_ID KEY_TOKENIZER_HF_JSON = Keys.Tokenizer.HF_JSON KEY_TOKENIZER_RWKV = Keys.Tokenizer.RWKV -KEY_TOKENIZER_PRIFIX_ID = Keys.Tokenizer.PREFIX_ID + +KEY_TOKENIZER_FIM_PRE_ID = Keys.Tokenizer.FIM_PRE_ID +KEY_TOKENIZER_FIM_SUF_ID = Keys.Tokenizer.FIM_SUF_ID +KEY_TOKENIZER_FIM_MID_ID = Keys.Tokenizer.FIM_MID_ID +KEY_TOKENIZER_FIM_PAD_ID = Keys.Tokenizer.FIM_PAD_ID +KEY_TOKENIZER_FIM_REP_ID = Keys.Tokenizer.FIM_REP_ID +KEY_TOKENIZER_FIM_SEP_ID = Keys.Tokenizer.FIM_SEP_ID + +# deprecated +KEY_TOKENIZER_PREFIX_ID = Keys.Tokenizer.PREFIX_ID KEY_TOKENIZER_SUFFIX_ID = Keys.Tokenizer.SUFFIX_ID KEY_TOKENIZER_MIDDLE_ID = Keys.Tokenizer.MIDDLE_ID -KEY_TOKENIZER_EOT_ID = Keys.Tokenizer.EOT_ID -KEY_TOKENIZER_EOM_ID = Keys.Tokenizer.EOM_ID diff --git a/gguf-py/gguf/gguf_writer.py b/gguf-py/gguf/gguf_writer.py index 5c460ef1b..0d8d8a0b0 100644 --- a/gguf-py/gguf/gguf_writer.py +++ b/gguf-py/gguf/gguf_writer.py @@ -843,15 +843,6 @@ class GGUFWriter: self.add_string(Keys.Tokenizer.CHAT_TEMPLATE, value) - def add_prefix_token_id(self, id: int) -> None: - self.add_uint32(Keys.Tokenizer.PREFIX_ID, id) - - def add_suffix_token_id(self, id: int) -> None: - self.add_uint32(Keys.Tokenizer.SUFFIX_ID, id) - - def add_middle_token_id(self, id: int) -> None: - self.add_uint32(Keys.Tokenizer.MIDDLE_ID, id) - def add_eot_token_id(self, id: int) -> None: self.add_uint32(Keys.Tokenizer.EOT_ID, id) diff --git a/include/llama.h b/include/llama.h index 4f8f6d23d..9110b5956 100644 --- a/include/llama.h +++ b/include/llama.h @@ -897,6 +897,7 @@ extern "C" { // Special tokens LLAMA_API llama_token llama_token_bos(const struct llama_model * model); // beginning-of-sentence LLAMA_API llama_token llama_token_eos(const struct llama_model * model); // end-of-sentence + LLAMA_API llama_token llama_token_eot(const struct llama_model * model); // end-of-turn LLAMA_API llama_token llama_token_cls(const struct llama_model * model); // classification LLAMA_API llama_token llama_token_sep(const struct llama_model * model); // sentence separator LLAMA_API llama_token llama_token_nl (const struct llama_model * model); // next-line @@ -905,11 +906,17 @@ extern "C" { LLAMA_API bool llama_add_bos_token(const struct llama_model * model); LLAMA_API bool llama_add_eos_token(const struct llama_model * model); - // Codellama infill tokens - LLAMA_API llama_token llama_token_prefix(const struct llama_model * model); // Beginning of infill prefix - LLAMA_API llama_token llama_token_middle(const struct llama_model * model); // Beginning of infill middle - LLAMA_API llama_token llama_token_suffix(const struct llama_model * model); // Beginning of infill suffix - LLAMA_API llama_token llama_token_eot (const struct llama_model * model); // End of infill middle + // infill tokens + DEPRECATED(LLAMA_API llama_token llama_token_prefix(const struct llama_model * model), "use llama_token_fim_pre instead"); + DEPRECATED(LLAMA_API llama_token llama_token_middle(const struct llama_model * model), "use llama_token_fim_mid instead"); + DEPRECATED(LLAMA_API llama_token llama_token_suffix(const struct llama_model * model), "use llama_token_fim_suf instead"); + + LLAMA_API llama_token llama_token_fim_pre(const struct llama_model * model); + LLAMA_API llama_token llama_token_fim_suf(const struct llama_model * model); + LLAMA_API llama_token llama_token_fim_mid(const struct llama_model * model); + LLAMA_API llama_token llama_token_fim_pad(const struct llama_model * model); + LLAMA_API llama_token llama_token_fim_rep(const struct llama_model * model); + LLAMA_API llama_token llama_token_fim_sep(const struct llama_model * model); // // Tokenization diff --git a/src/llama-vocab.cpp b/src/llama-vocab.cpp index d2f34ddd6..a27394a37 100644 --- a/src/llama-vocab.cpp +++ b/src/llama-vocab.cpp @@ -1663,6 +1663,14 @@ llama_token llama_token_eos_impl(const struct llama_vocab & vocab) { return vocab.special_eos_id; } +llama_token llama_token_eot_impl(const struct llama_vocab & vocab) { + return vocab.special_eot_id; +} + +llama_token llama_token_eom_impl(const struct llama_vocab & vocab) { + return vocab.special_eom_id; +} + llama_token llama_token_cls_impl(const struct llama_vocab & vocab) { return vocab.special_cls_id; } @@ -1688,23 +1696,39 @@ bool llama_add_eos_token_impl(const struct llama_vocab & vocab) { } llama_token llama_token_prefix_impl(const struct llama_vocab & vocab) { - return vocab.special_prefix_id; + return vocab.special_fim_pre_id; } llama_token llama_token_middle_impl(const struct llama_vocab & vocab) { - return vocab.special_middle_id; + return vocab.special_fim_mid_id; } llama_token llama_token_suffix_impl(const struct llama_vocab & vocab) { - return vocab.special_suffix_id; + return vocab.special_fim_suf_id; } -llama_token llama_token_eot_impl(const struct llama_vocab & vocab) { - return vocab.special_eot_id; +llama_token llama_token_fim_pre_impl(const struct llama_vocab & vocab) { + return vocab.special_fim_pre_id; } -llama_token llama_token_eom_impl(const struct llama_vocab & vocab) { - return vocab.special_eom_id; +llama_token llama_token_fim_suf_impl(const struct llama_vocab & vocab) { + return vocab.special_fim_suf_id; +} + +llama_token llama_token_fim_mid_impl(const struct llama_vocab & vocab) { + return vocab.special_fim_mid_id; +} + +llama_token llama_token_fim_pad_impl(const struct llama_vocab & vocab) { + return vocab.special_fim_pad_id; +} + +llama_token llama_token_fim_rep_impl(const struct llama_vocab & vocab) { + return vocab.special_fim_rep_id; +} + +llama_token llama_token_fim_sep_impl(const struct llama_vocab & vocab) { + return vocab.special_fim_sep_id; } int32_t llama_tokenize_impl( diff --git a/src/llama-vocab.h b/src/llama-vocab.h index 28bad9135..17e14488a 100644 --- a/src/llama-vocab.h +++ b/src/llama-vocab.h @@ -37,20 +37,26 @@ struct llama_vocab { std::map, int> bpe_ranks; // default LLaMA special tokens + // TODO: should we set all of these to LLAMA_TOKEN_NULL? id special_bos_id = 1; id special_eos_id = 2; + id special_eot_id = LLAMA_TOKEN_NULL; + id special_eom_id = LLAMA_TOKEN_NULL; id special_unk_id = 0; id special_sep_id = LLAMA_TOKEN_NULL; id special_pad_id = LLAMA_TOKEN_NULL; id special_cls_id = LLAMA_TOKEN_NULL; id special_mask_id = LLAMA_TOKEN_NULL; - id linefeed_id = 13; - id special_prefix_id = LLAMA_TOKEN_NULL; - id special_suffix_id = LLAMA_TOKEN_NULL; - id special_middle_id = LLAMA_TOKEN_NULL; - id special_eot_id = LLAMA_TOKEN_NULL; // TODO: move above after "eos_id", and here add "file separator" token - id special_eom_id = LLAMA_TOKEN_NULL; + id linefeed_id = 13; + + // fim tokens + id special_fim_pre_id = LLAMA_TOKEN_NULL; + id special_fim_suf_id = LLAMA_TOKEN_NULL; + id special_fim_mid_id = LLAMA_TOKEN_NULL; + id special_fim_pad_id = LLAMA_TOKEN_NULL; + id special_fim_rep_id = LLAMA_TOKEN_NULL; // repo + id special_fim_sep_id = LLAMA_TOKEN_NULL; // file separator // set of all tokens that cause "end of generation" std::set special_eog_ids; @@ -104,19 +110,26 @@ bool llama_token_is_control_impl(const struct llama_vocab & vocab, llama_token t llama_token llama_token_bos_impl(const struct llama_vocab & vocab); llama_token llama_token_eos_impl(const struct llama_vocab & vocab); +llama_token llama_token_eot_impl(const struct llama_vocab & vocab); +llama_token llama_token_eom_impl(const struct llama_vocab & vocab); llama_token llama_token_cls_impl(const struct llama_vocab & vocab); llama_token llama_token_sep_impl(const struct llama_vocab & vocab); llama_token llama_token_nl_impl (const struct llama_vocab & vocab); llama_token llama_token_pad_impl(const struct llama_vocab & vocab); -bool llama_add_bos_token_impl(const struct llama_vocab & vocab); -bool llama_add_eos_token_impl(const struct llama_vocab & vocab); - llama_token llama_token_prefix_impl(const struct llama_vocab & vocab); llama_token llama_token_middle_impl(const struct llama_vocab & vocab); llama_token llama_token_suffix_impl(const struct llama_vocab & vocab); -llama_token llama_token_eot_impl (const struct llama_vocab & vocab); -llama_token llama_token_eom_impl (const struct llama_vocab & vocab); + +llama_token llama_token_fim_pre_impl(const struct llama_vocab & vocab); +llama_token llama_token_fim_suf_impl(const struct llama_vocab & vocab); +llama_token llama_token_fim_mid_impl(const struct llama_vocab & vocab); +llama_token llama_token_fim_pad_impl(const struct llama_vocab & vocab); +llama_token llama_token_fim_rep_impl(const struct llama_vocab & vocab); +llama_token llama_token_fim_sep_impl(const struct llama_vocab & vocab); + +bool llama_add_bos_token_impl(const struct llama_vocab & vocab); +bool llama_add_eos_token_impl(const struct llama_vocab & vocab); int32_t llama_tokenize_impl( const struct llama_vocab & vocab, diff --git a/src/llama.cpp b/src/llama.cpp index da7afb1ee..f68024f5b 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -345,6 +345,8 @@ enum llm_kv { LLM_KV_TOKENIZER_MERGES, LLM_KV_TOKENIZER_BOS_ID, LLM_KV_TOKENIZER_EOS_ID, + LLM_KV_TOKENIZER_EOT_ID, + LLM_KV_TOKENIZER_EOM_ID, LLM_KV_TOKENIZER_UNK_ID, LLM_KV_TOKENIZER_SEP_ID, LLM_KV_TOKENIZER_PAD_ID, @@ -357,14 +359,20 @@ enum llm_kv { LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, LLM_KV_TOKENIZER_HF_JSON, LLM_KV_TOKENIZER_RWKV, - LLM_KV_TOKENIZER_PREFIX_ID, - LLM_KV_TOKENIZER_SUFFIX_ID, - LLM_KV_TOKENIZER_MIDDLE_ID, - LLM_KV_TOKENIZER_EOT_ID, - LLM_KV_TOKENIZER_EOM_ID, + LLM_KV_TOKENIZER_FIM_PRE_ID, + LLM_KV_TOKENIZER_FIM_SUF_ID, + LLM_KV_TOKENIZER_FIM_MID_ID, + LLM_KV_TOKENIZER_FIM_PAD_ID, + LLM_KV_TOKENIZER_FIM_REP_ID, + LLM_KV_TOKENIZER_FIM_SEP_ID, LLM_KV_ADAPTER_TYPE, LLM_KV_ADAPTER_LORA_ALPHA, + + // deprecated: + LLM_KV_TOKENIZER_PREFIX_ID, + LLM_KV_TOKENIZER_SUFFIX_ID, + LLM_KV_TOKENIZER_MIDDLE_ID, }; static const std::map LLM_KV_NAMES = { @@ -422,57 +430,65 @@ static const std::map LLM_KV_NAMES = { { LLM_KV_ATTENTION_SLIDING_WINDOW, "%s.attention.sliding_window" }, { LLM_KV_ATTENTION_SCALE, "%s.attention.scale" }, - { LLM_KV_ROPE_DIMENSION_COUNT, "%s.rope.dimension_count" }, - { LLM_KV_ROPE_FREQ_BASE, "%s.rope.freq_base" }, - { LLM_KV_ROPE_SCALE_LINEAR, "%s.rope.scale_linear" }, - { LLM_KV_ROPE_SCALING_TYPE, "%s.rope.scaling.type" }, - { LLM_KV_ROPE_SCALING_FACTOR, "%s.rope.scaling.factor" }, - { LLM_KV_ROPE_SCALING_ATTN_FACTOR, "%s.rope.scaling.attn_factor" }, - { LLM_KV_ROPE_SCALING_ORIG_CTX_LEN, "%s.rope.scaling.original_context_length" }, - { LLM_KV_ROPE_SCALING_FINETUNED, "%s.rope.scaling.finetuned" }, - { LLM_KV_ROPE_SCALING_YARN_LOG_MUL, "%s.rope.scaling.yarn_log_multiplier" }, + { LLM_KV_ROPE_DIMENSION_COUNT, "%s.rope.dimension_count" }, + { LLM_KV_ROPE_FREQ_BASE, "%s.rope.freq_base" }, + { LLM_KV_ROPE_SCALE_LINEAR, "%s.rope.scale_linear" }, + { LLM_KV_ROPE_SCALING_TYPE, "%s.rope.scaling.type" }, + { LLM_KV_ROPE_SCALING_FACTOR, "%s.rope.scaling.factor" }, + { LLM_KV_ROPE_SCALING_ATTN_FACTOR, "%s.rope.scaling.attn_factor" }, + { LLM_KV_ROPE_SCALING_ORIG_CTX_LEN, "%s.rope.scaling.original_context_length" }, + { LLM_KV_ROPE_SCALING_FINETUNED, "%s.rope.scaling.finetuned" }, + { LLM_KV_ROPE_SCALING_YARN_LOG_MUL, "%s.rope.scaling.yarn_log_multiplier" }, - { LLM_KV_SPLIT_NO, "split.no" }, - { LLM_KV_SPLIT_COUNT, "split.count" }, - { LLM_KV_SPLIT_TENSORS_COUNT, "split.tensors.count" }, + { LLM_KV_SPLIT_NO, "split.no" }, + { LLM_KV_SPLIT_COUNT, "split.count" }, + { LLM_KV_SPLIT_TENSORS_COUNT, "split.tensors.count" }, - { LLM_KV_SSM_CONV_KERNEL, "%s.ssm.conv_kernel" }, - { LLM_KV_SSM_INNER_SIZE, "%s.ssm.inner_size" }, - { LLM_KV_SSM_STATE_SIZE, "%s.ssm.state_size" }, - { LLM_KV_SSM_TIME_STEP_RANK, "%s.ssm.time_step_rank" }, - { LLM_KV_SSM_DT_B_C_RMS, "%s.ssm.dt_b_c_rms" }, + { LLM_KV_SSM_CONV_KERNEL, "%s.ssm.conv_kernel" }, + { LLM_KV_SSM_INNER_SIZE, "%s.ssm.inner_size" }, + { LLM_KV_SSM_STATE_SIZE, "%s.ssm.state_size" }, + { LLM_KV_SSM_TIME_STEP_RANK, "%s.ssm.time_step_rank" }, + { LLM_KV_SSM_DT_B_C_RMS, "%s.ssm.dt_b_c_rms" }, - { LLM_KV_WKV_HEAD_SIZE, "%s.wkv.head_size" }, + { LLM_KV_WKV_HEAD_SIZE, "%s.wkv.head_size" }, - { LLM_KV_TOKENIZER_MODEL, "tokenizer.ggml.model" }, - { LLM_KV_TOKENIZER_PRE, "tokenizer.ggml.pre" }, - { LLM_KV_TOKENIZER_LIST, "tokenizer.ggml.tokens" }, - { LLM_KV_TOKENIZER_TOKEN_TYPE, "tokenizer.ggml.token_type" }, - { LLM_KV_TOKENIZER_TOKEN_TYPE_COUNT, "tokenizer.ggml.token_type_count" }, - { LLM_KV_TOKENIZER_SCORES, "tokenizer.ggml.scores" }, - { LLM_KV_TOKENIZER_MERGES, "tokenizer.ggml.merges" }, - { LLM_KV_TOKENIZER_BOS_ID, "tokenizer.ggml.bos_token_id" }, - { LLM_KV_TOKENIZER_EOS_ID, "tokenizer.ggml.eos_token_id" }, - { LLM_KV_TOKENIZER_UNK_ID, "tokenizer.ggml.unknown_token_id" }, - { LLM_KV_TOKENIZER_SEP_ID, "tokenizer.ggml.seperator_token_id" }, - { LLM_KV_TOKENIZER_PAD_ID, "tokenizer.ggml.padding_token_id" }, - { LLM_KV_TOKENIZER_CLS_ID, "tokenizer.ggml.cls_token_id" }, - { LLM_KV_TOKENIZER_MASK_ID, "tokenizer.ggml.mask_token_id" }, - { LLM_KV_TOKENIZER_ADD_BOS, "tokenizer.ggml.add_bos_token" }, - { LLM_KV_TOKENIZER_ADD_EOS, "tokenizer.ggml.add_eos_token" }, - { LLM_KV_TOKENIZER_ADD_PREFIX, "tokenizer.ggml.add_space_prefix" }, - { LLM_KV_TOKENIZER_REMOVE_EXTRA_WS, "tokenizer.ggml.remove_extra_whitespaces" }, - { LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, "tokenizer.ggml.precompiled_charsmap" }, - { LLM_KV_TOKENIZER_HF_JSON, "tokenizer.huggingface.json" }, - { LLM_KV_TOKENIZER_RWKV, "tokenizer.rwkv.world" }, - { LLM_KV_TOKENIZER_PREFIX_ID, "tokenizer.ggml.prefix_token_id" }, - { LLM_KV_TOKENIZER_SUFFIX_ID, "tokenizer.ggml.suffix_token_id" }, - { LLM_KV_TOKENIZER_MIDDLE_ID, "tokenizer.ggml.middle_token_id" }, - { LLM_KV_TOKENIZER_EOT_ID, "tokenizer.ggml.eot_token_id" }, - { LLM_KV_TOKENIZER_EOM_ID, "tokenizer.ggml.eom_token_id" }, + { LLM_KV_TOKENIZER_MODEL, "tokenizer.ggml.model" }, + { LLM_KV_TOKENIZER_PRE, "tokenizer.ggml.pre" }, + { LLM_KV_TOKENIZER_LIST, "tokenizer.ggml.tokens" }, + { LLM_KV_TOKENIZER_TOKEN_TYPE, "tokenizer.ggml.token_type" }, + { LLM_KV_TOKENIZER_TOKEN_TYPE_COUNT, "tokenizer.ggml.token_type_count" }, + { LLM_KV_TOKENIZER_SCORES, "tokenizer.ggml.scores" }, + { LLM_KV_TOKENIZER_MERGES, "tokenizer.ggml.merges" }, + { LLM_KV_TOKENIZER_BOS_ID, "tokenizer.ggml.bos_token_id" }, + { LLM_KV_TOKENIZER_EOS_ID, "tokenizer.ggml.eos_token_id" }, + { LLM_KV_TOKENIZER_EOT_ID, "tokenizer.ggml.eot_token_id" }, + { LLM_KV_TOKENIZER_EOM_ID, "tokenizer.ggml.eom_token_id" }, + { LLM_KV_TOKENIZER_UNK_ID, "tokenizer.ggml.unknown_token_id" }, + { LLM_KV_TOKENIZER_SEP_ID, "tokenizer.ggml.seperator_token_id" }, + { LLM_KV_TOKENIZER_PAD_ID, "tokenizer.ggml.padding_token_id" }, + { LLM_KV_TOKENIZER_CLS_ID, "tokenizer.ggml.cls_token_id" }, + { LLM_KV_TOKENIZER_MASK_ID, "tokenizer.ggml.mask_token_id" }, + { LLM_KV_TOKENIZER_ADD_BOS, "tokenizer.ggml.add_bos_token" }, + { LLM_KV_TOKENIZER_ADD_EOS, "tokenizer.ggml.add_eos_token" }, + { LLM_KV_TOKENIZER_ADD_PREFIX, "tokenizer.ggml.add_space_prefix" }, + { LLM_KV_TOKENIZER_REMOVE_EXTRA_WS, "tokenizer.ggml.remove_extra_whitespaces" }, + { LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, "tokenizer.ggml.precompiled_charsmap" }, + { LLM_KV_TOKENIZER_HF_JSON, "tokenizer.huggingface.json" }, + { LLM_KV_TOKENIZER_RWKV, "tokenizer.rwkv.world" }, + { LLM_KV_TOKENIZER_FIM_PRE_ID, "tokenizer.ggml.fim_pre_token_id" }, + { LLM_KV_TOKENIZER_FIM_SUF_ID, "tokenizer.ggml.fim_suf_token_id" }, + { LLM_KV_TOKENIZER_FIM_MID_ID, "tokenizer.ggml.fim_mid_token_id" }, + { LLM_KV_TOKENIZER_FIM_PAD_ID, "tokenizer.ggml.fim_pad_token_id" }, + { LLM_KV_TOKENIZER_FIM_REP_ID, "tokenizer.ggml.fim_rep_token_id" }, + { LLM_KV_TOKENIZER_FIM_SEP_ID, "tokenizer.ggml.fim_sep_token_id" }, - { LLM_KV_ADAPTER_TYPE, "adapter.type" }, - { LLM_KV_ADAPTER_LORA_ALPHA, "adapter.lora.alpha" }, + { LLM_KV_ADAPTER_TYPE, "adapter.type" }, + { LLM_KV_ADAPTER_LORA_ALPHA, "adapter.lora.alpha" }, + + // deprecated + { LLM_KV_TOKENIZER_PREFIX_ID, "tokenizer.ggml.prefix_token_id" }, + { LLM_KV_TOKENIZER_SUFFIX_ID, "tokenizer.ggml.suffix_token_id" }, + { LLM_KV_TOKENIZER_MIDDLE_ID, "tokenizer.ggml.middle_token_id" }, }; struct LLM_KV { @@ -6164,14 +6180,14 @@ static void llm_load_vocab( vocab.type = LLAMA_VOCAB_TYPE_NONE; // default special tokens - vocab.special_bos_id = -1; - vocab.special_eos_id = -1; - vocab.special_unk_id = -1; - vocab.special_sep_id = -1; - vocab.special_pad_id = -1; - vocab.special_cls_id = -1; - vocab.special_mask_id = -1; - vocab.linefeed_id = -1; + vocab.special_bos_id = LLAMA_TOKEN_NULL; + vocab.special_eos_id = LLAMA_TOKEN_NULL; + vocab.special_unk_id = LLAMA_TOKEN_NULL; + vocab.special_sep_id = LLAMA_TOKEN_NULL; + vocab.special_pad_id = LLAMA_TOKEN_NULL; + vocab.special_cls_id = LLAMA_TOKEN_NULL; + vocab.special_mask_id = LLAMA_TOKEN_NULL; + vocab.linefeed_id = LLAMA_TOKEN_NULL; // read vocab size from metadata if (!ml.get_key(LLM_KV_VOCAB_SIZE, vocab.n_vocab, false)) { @@ -6188,16 +6204,16 @@ static void llm_load_vocab( vocab.special_bos_id = 1; vocab.special_eos_id = 2; vocab.special_unk_id = 0; - vocab.special_sep_id = -1; - vocab.special_pad_id = -1; - vocab.special_cls_id = -1; - vocab.special_mask_id = -1; + vocab.special_sep_id = LLAMA_TOKEN_NULL; + vocab.special_pad_id = LLAMA_TOKEN_NULL; + vocab.special_cls_id = LLAMA_TOKEN_NULL; + vocab.special_mask_id = LLAMA_TOKEN_NULL; } else if (tokenizer_model == "bert") { vocab.type = LLAMA_VOCAB_TYPE_WPM; // default special tokens - vocab.special_bos_id = -1; - vocab.special_eos_id = -1; + vocab.special_bos_id = LLAMA_TOKEN_NULL; + vocab.special_eos_id = LLAMA_TOKEN_NULL; vocab.special_unk_id = 100; vocab.special_sep_id = 102; vocab.special_pad_id = 0; @@ -6233,22 +6249,22 @@ static void llm_load_vocab( // default special tokens vocab.special_bos_id = 11; vocab.special_eos_id = 11; - vocab.special_unk_id = -1; - vocab.special_sep_id = -1; - vocab.special_pad_id = -1; - vocab.special_cls_id = -1; - vocab.special_mask_id = -1; + vocab.special_unk_id = LLAMA_TOKEN_NULL; + vocab.special_sep_id = LLAMA_TOKEN_NULL; + vocab.special_pad_id = LLAMA_TOKEN_NULL; + vocab.special_cls_id = LLAMA_TOKEN_NULL; + vocab.special_mask_id = LLAMA_TOKEN_NULL; } else if (tokenizer_model == "t5") { vocab.type = LLAMA_VOCAB_TYPE_UGM; // default special tokens - vocab.special_bos_id = -1; + vocab.special_bos_id = LLAMA_TOKEN_NULL; vocab.special_eos_id = 1; vocab.special_unk_id = 2; - vocab.special_sep_id = -1; + vocab.special_sep_id = LLAMA_TOKEN_NULL; vocab.special_pad_id = 0; - vocab.special_cls_id = -1; - vocab.special_mask_id = -1; + vocab.special_cls_id = LLAMA_TOKEN_NULL; + vocab.special_mask_id = LLAMA_TOKEN_NULL; const int precompiled_charsmap_keyidx = gguf_find_key(ctx, kv(LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP).c_str()); if (precompiled_charsmap_keyidx != -1) { @@ -6271,11 +6287,11 @@ static void llm_load_vocab( vocab.type = LLAMA_VOCAB_TYPE_RWKV; // default special tokens - vocab.special_bos_id = -1; - vocab.special_eos_id = -1; - vocab.special_unk_id = -1; - vocab.special_sep_id = -1; - vocab.special_pad_id = -1; + vocab.special_bos_id = LLAMA_TOKEN_NULL; + vocab.special_eos_id = LLAMA_TOKEN_NULL; + vocab.special_unk_id = LLAMA_TOKEN_NULL; + vocab.special_sep_id = LLAMA_TOKEN_NULL; + vocab.special_pad_id = LLAMA_TOKEN_NULL; } else { throw std::runtime_error(format("unknown tokenizer: '%s'", tokenizer_model.c_str())); } @@ -6359,7 +6375,7 @@ static void llm_load_vocab( } else if ( tokenizer_pre == "chatglm-bpe") { vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_CHATGLM4; - vocab.special_bos_id = -1; + vocab.special_bos_id = LLAMA_TOKEN_NULL; } else if ( tokenizer_pre == "viking") { vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_VIKING; @@ -6485,44 +6501,6 @@ static void llm_load_vocab( // determine the newline token: LLaMA "<0x0A>" == 10 == '\n', Falcon 193 == '\n' if (vocab.type == LLAMA_VOCAB_TYPE_SPM) { - // For Fill-In-the-Middle (FIM)/infill models which where converted - // prior to support of FIM special tokens in GGUF, the following - // will allow those models to continue to work. The general names - // of the known models are currently CodeLlama (LLM_ARCH_LLAMA) and - // CodeGemma (LLM_ARCH_GEMMA). This can potentially be removed once - // new versions of these models have been published. - std::string gen_name; - ml.get_key(LLM_KV_GENERAL_NAME, gen_name, false); - - std::transform(gen_name.begin(), gen_name.end(), gen_name.begin(), - [](unsigned char c){ return std::tolower(c); }); - - if (gen_name.find("code") != std::string::npos) { - if (model.arch == LLM_ARCH_LLAMA - && 32010 < vocab.id_to_token.size() - && vocab.id_to_token[32007].text.find("
") != std::string::npos
-              && vocab.id_to_token[32008].text.find("") != std::string::npos
-              && vocab.id_to_token[32009].text.find("") != std::string::npos
-              && vocab.id_to_token[32010].text.find("") != std::string::npos) {
-                vocab.special_prefix_id = 32007;
-                vocab.special_suffix_id = 32008;
-                vocab.special_middle_id = 32009;
-                vocab.special_eot_id    = 32010;
-            } else if (model.arch == LLM_ARCH_GEMMA
-              && 107 < vocab.id_to_token.size()
-              && vocab.id_to_token[67].text == "<|fim_prefix|>"
-              && vocab.id_to_token[69].text == "<|fim_suffix|>"
-              && vocab.id_to_token[68].text == "<|fim_middle|>"
-              && vocab.id_to_token[107].text == "") {
-                vocab.special_prefix_id = 67;
-                vocab.special_suffix_id = 69;
-                vocab.special_middle_id = 68;
-                // TODO: this is not EOT, it is "file separator" token, needs fix
-                //       https://huggingface.co/google/codegemma-7b-it/blob/9b1d9231388358c04d90bd003458f5070d97db44/tokenizer_config.json#L565-L572
-                //vocab.special_eot_id    = 70;
-                vocab.special_eot_id    = 107;
-            }
-        }
         try {
             vocab.linefeed_id = llama_byte_to_token_impl(vocab, '\n');
         } catch (const std::exception & e) {
@@ -6550,18 +6528,26 @@ static void llm_load_vocab(
     // special tokens
     {
         const std::vector> special_token_types = {
-            { LLM_KV_TOKENIZER_BOS_ID,    vocab.special_bos_id    },
-            { LLM_KV_TOKENIZER_EOS_ID,    vocab.special_eos_id    },
-            { LLM_KV_TOKENIZER_UNK_ID,    vocab.special_unk_id    },
-            { LLM_KV_TOKENIZER_SEP_ID,    vocab.special_sep_id    },
-            { LLM_KV_TOKENIZER_PAD_ID,    vocab.special_pad_id    },
-            { LLM_KV_TOKENIZER_CLS_ID,    vocab.special_cls_id    },
-            { LLM_KV_TOKENIZER_MASK_ID,   vocab.special_mask_id   },
-            { LLM_KV_TOKENIZER_PREFIX_ID, vocab.special_prefix_id },
-            { LLM_KV_TOKENIZER_SUFFIX_ID, vocab.special_suffix_id },
-            { LLM_KV_TOKENIZER_MIDDLE_ID, vocab.special_middle_id },
-            { LLM_KV_TOKENIZER_EOT_ID,    vocab.special_eot_id    },
-            { LLM_KV_TOKENIZER_EOM_ID,    vocab.special_eom_id    },
+            { LLM_KV_TOKENIZER_BOS_ID,     vocab.special_bos_id     },
+            { LLM_KV_TOKENIZER_EOS_ID,     vocab.special_eos_id     },
+            { LLM_KV_TOKENIZER_EOT_ID,     vocab.special_eot_id     },
+            { LLM_KV_TOKENIZER_EOM_ID,     vocab.special_eom_id     },
+            { LLM_KV_TOKENIZER_UNK_ID,     vocab.special_unk_id     },
+            { LLM_KV_TOKENIZER_SEP_ID,     vocab.special_sep_id     },
+            { LLM_KV_TOKENIZER_PAD_ID,     vocab.special_pad_id     },
+            { LLM_KV_TOKENIZER_CLS_ID,     vocab.special_cls_id     },
+            { LLM_KV_TOKENIZER_MASK_ID,    vocab.special_mask_id    },
+            { LLM_KV_TOKENIZER_FIM_PRE_ID, vocab.special_fim_pre_id },
+            { LLM_KV_TOKENIZER_FIM_SUF_ID, vocab.special_fim_suf_id },
+            { LLM_KV_TOKENIZER_FIM_MID_ID, vocab.special_fim_mid_id },
+            { LLM_KV_TOKENIZER_FIM_PAD_ID, vocab.special_fim_pad_id },
+            { LLM_KV_TOKENIZER_FIM_REP_ID, vocab.special_fim_rep_id },
+            { LLM_KV_TOKENIZER_FIM_SEP_ID, vocab.special_fim_sep_id },
+
+            // deprecated
+            { LLM_KV_TOKENIZER_PREFIX_ID, vocab.special_fim_pre_id },
+            { LLM_KV_TOKENIZER_SUFFIX_ID, vocab.special_fim_suf_id },
+            { LLM_KV_TOKENIZER_MIDDLE_ID, vocab.special_fim_mid_id },
         };
 
         for (const auto & it : special_token_types) {
@@ -6592,22 +6578,21 @@ static void llm_load_vocab(
             }
         }
 
-        // find EOT token: "<|eot_id|>", "<|im_end|>", "", etc.
-        //
-        // TODO: convert scripts should provide this token through the KV metadata LLAMA_KV_TOKENIZER_EOT_ID
-        //       for now, we apply this workaround to find the EOT token based on its text
-        if (vocab.special_eot_id == -1) {
-            for (const auto & t : vocab.token_to_id) {
+        // auto-detect special tokens by text
+        // TODO: convert scripts should provide these tokens through the KV metadata LLM_KV_TOKENIZER_...
+        //       for now, we apply this workaround to find the tokens based on their text
+
+        for (const auto & t : vocab.token_to_id) {
+            // find EOT token: "<|eot_id|>", "<|im_end|>", "", etc.
+            if (vocab.special_eot_id == LLAMA_TOKEN_NULL) {
                 if (false
-                        // TODO: gemma "" is exported as a normal token, so the following check does not work
-                        //       need to fix convert script
-                        //vocab.id_to_token[t.second].type == LLAMA_TOKEN_TYPE_CONTROL &&
                         || t.first == "<|eot_id|>"
                         || t.first == "<|im_end|>"
                         || t.first == "<|end|>"
                         || t.first == ""
                         || t.first == "<|endoftext|>"
                         || t.first == ""
+                        || t.first == "<｜end▁of▁sentence｜>" // DeepSeek
                    ) {
                     vocab.special_eot_id = t.second;
                     if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
@@ -6615,23 +6600,118 @@ static void llm_load_vocab(
                                 __func__, t.first.c_str());
                         vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                     }
-                    break;
                 }
             }
-        }
 
-        // find EOM token: "<|eom_id|>"
-        //
-        // TODO: convert scripts should provide this token through the KV metadata LLAMA_KV_TOKENIZER_EOM_ID
-        //       for now, we apply this workaround to find the EOM token based on its text
-        if (vocab.special_eom_id == -1) {
-            const auto & t = vocab.token_to_id.find("<|eom_id|>");
-            if (t != vocab.token_to_id.end()) {
-                vocab.special_eom_id = t->second;
-                if ((vocab.id_to_token[t->second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                    LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                        __func__, t->first.c_str());
-                    vocab.id_to_token[t->second].attr = LLAMA_TOKEN_ATTR_CONTROL;
+            // find EOM token: "<|eom_id|>"
+            if (vocab.special_eom_id == LLAMA_TOKEN_NULL) {
+                if (false
+                        || t.first == "<|eom_id|>"
+                        ) {
+                    vocab.special_eom_id = t.second;
+                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
+                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.first.c_str());
+                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
+                    }
+                }
+            }
+
+            // find FIM_PRE token: "<|fim_prefix|>", "", "
", etc.
+            if (vocab.special_fim_pre_id == LLAMA_TOKEN_NULL) {
+                if (false
+                        || t.first == "<|fim_prefix|>"  // Qwen
+                        || t.first == ""
+                        || t.first == "<｜fim▁begin｜>" // DeepSeek
+                        || t.first == "
"
+                        ) {
+                    vocab.special_fim_pre_id = t.second;
+                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
+                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.first.c_str());
+                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
+                    }
+                }
+            }
+
+            // find FIM_SUF token: "<|fim_suffix|>", "", "", etc.
+            if (vocab.special_fim_suf_id == LLAMA_TOKEN_NULL) {
+                if (false
+                        || t.first == "<|fim_suffix|>" // Qwen
+                        || t.first == ""
+                        || t.first == "<｜fim▁hole｜>" // DeepSeek
+                        || t.first == ""
+                        ) {
+                    vocab.special_fim_suf_id = t.second;
+                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
+                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.first.c_str());
+                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
+                    }
+                }
+            }
+
+            // find FIM_MID token: "<|fim_middle|>", "", "", etc.
+            if (vocab.special_fim_mid_id == LLAMA_TOKEN_NULL) {
+                if (false
+                        || t.first == "<|fim_middle|>" // Qwen
+                        || t.first == ""
+                        || t.first == "<｜fim▁end｜>"  // DeepSeek
+                        || t.first == ""
+                        ) {
+                    vocab.special_fim_mid_id = t.second;
+                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
+                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.first.c_str());
+                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
+                    }
+                }
+            }
+
+            // find FIM_PAD token: "<|fim_pad|>", "", "", etc.
+            if (vocab.special_fim_pad_id == LLAMA_TOKEN_NULL) {
+                if (false
+                        || t.first == "<|fim_pad|>" // Qwen
+                        || t.first == ""
+                        || t.first == ""
+                        ) {
+                    vocab.special_fim_pad_id = t.second;
+                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
+                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.first.c_str());
+                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
+                    }
+                }
+            }
+
+            // find FIM_REP token: "<|fim_repo|>", "", "", etc.
+            if (vocab.special_fim_rep_id == LLAMA_TOKEN_NULL) {
+                if (false
+                        || t.first == "<|fim_repo|>"  // Qwen
+                        || t.first == "<|repo_name|>"
+                        || t.first == ""
+                        || t.first == ""
+                        ) {
+                    vocab.special_fim_rep_id = t.second;
+                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
+                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.first.c_str());
+                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
+                    }
+                }
+            }
+
+            // find FIM_SEP token: "<|file_sep|>"
+            if (vocab.special_fim_sep_id == LLAMA_TOKEN_NULL) {
+                if (false
+                        || t.first == "<|file_sep|>" // Qwen
+                        ) {
+                    vocab.special_fim_sep_id = t.second;
+                    if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
+                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.first.c_str());
+                        vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
+                    }
                 }
             }
         }
@@ -6659,17 +6739,17 @@ static void llm_load_vocab(
             }
         }
 
-        if (vocab.special_eos_id != -1 && vocab.special_eog_ids.count(vocab.special_eos_id) == 0) {
+        if (vocab.special_eos_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_eos_id) == 0) {
             vocab.special_eog_ids.insert(vocab.special_eos_id);
             LLAMA_LOG_WARN("%s: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);
         }
 
-        if (vocab.special_eot_id != -1 && vocab.special_eog_ids.count(vocab.special_eot_id) == 0) {
+        if (vocab.special_eot_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_eot_id) == 0) {
             vocab.special_eog_ids.insert(vocab.special_eot_id);
             LLAMA_LOG_WARN("%s: special_eot_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);
         }
 
-        if (vocab.special_eom_id != -1 && vocab.special_eog_ids.count(vocab.special_eom_id) == 0) {
+        if (vocab.special_eom_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_eom_id) == 0) {
             vocab.special_eog_ids.insert(vocab.special_eom_id);
             LLAMA_LOG_WARN("%s: special_eom_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);
         }
@@ -6863,20 +6943,24 @@ static void llm_load_print_meta(llama_model_loader & ml, llama_model & model) {
     LLAMA_LOG_INFO("%s: general.name     = %s\n",    __func__, model.name.c_str());
 
     // special tokens
-    if (vocab.special_bos_id    != -1) { LLAMA_LOG_INFO( "%s: BOS token        = %d '%s'\n", __func__, vocab.special_bos_id,  vocab.id_to_token[vocab.special_bos_id].text.c_str() );  }
-    if (vocab.special_eos_id    != -1) { LLAMA_LOG_INFO( "%s: EOS token        = %d '%s'\n", __func__, vocab.special_eos_id,  vocab.id_to_token[vocab.special_eos_id].text.c_str() );  }
-    if (vocab.special_unk_id    != -1) { LLAMA_LOG_INFO( "%s: UNK token        = %d '%s'\n", __func__, vocab.special_unk_id,  vocab.id_to_token[vocab.special_unk_id].text.c_str() );  }
-    if (vocab.special_sep_id    != -1) { LLAMA_LOG_INFO( "%s: SEP token        = %d '%s'\n", __func__, vocab.special_sep_id,  vocab.id_to_token[vocab.special_sep_id].text.c_str() );  }
-    if (vocab.special_pad_id    != -1) { LLAMA_LOG_INFO( "%s: PAD token        = %d '%s'\n", __func__, vocab.special_pad_id,  vocab.id_to_token[vocab.special_pad_id].text.c_str() );  }
-    if (vocab.special_cls_id    != -1) { LLAMA_LOG_INFO( "%s: CLS token        = %d '%s'\n", __func__, vocab.special_cls_id,  vocab.id_to_token[vocab.special_cls_id].text.c_str() );  }
-    if (vocab.special_mask_id   != -1) { LLAMA_LOG_INFO( "%s: MASK token       = %d '%s'\n", __func__, vocab.special_mask_id, vocab.id_to_token[vocab.special_mask_id].text.c_str() ); }
+    if (vocab.special_bos_id  != -1)    { LLAMA_LOG_INFO( "%s: BOS token        = %d '%s'\n", __func__, vocab.special_bos_id,     vocab.id_to_token[vocab.special_bos_id].text.c_str() );  }
+    if (vocab.special_eos_id  != -1)    { LLAMA_LOG_INFO( "%s: EOS token        = %d '%s'\n", __func__, vocab.special_eos_id,     vocab.id_to_token[vocab.special_eos_id].text.c_str() );  }
+    if (vocab.special_eot_id  != -1)    { LLAMA_LOG_INFO( "%s: EOT token        = %d '%s'\n", __func__, vocab.special_eot_id,     vocab.id_to_token[vocab.special_eot_id].text.c_str() );  }
+    if (vocab.special_eom_id  != -1)    { LLAMA_LOG_INFO( "%s: EOM token        = %d '%s'\n", __func__, vocab.special_eom_id,     vocab.id_to_token[vocab.special_eom_id].text.c_str() );  }
+    if (vocab.special_unk_id  != -1)    { LLAMA_LOG_INFO( "%s: UNK token        = %d '%s'\n", __func__, vocab.special_unk_id,     vocab.id_to_token[vocab.special_unk_id].text.c_str() );  }
+    if (vocab.special_sep_id  != -1)    { LLAMA_LOG_INFO( "%s: SEP token        = %d '%s'\n", __func__, vocab.special_sep_id,     vocab.id_to_token[vocab.special_sep_id].text.c_str() );  }
+    if (vocab.special_pad_id  != -1)    { LLAMA_LOG_INFO( "%s: PAD token        = %d '%s'\n", __func__, vocab.special_pad_id,     vocab.id_to_token[vocab.special_pad_id].text.c_str() );  }
+    if (vocab.special_cls_id  != -1)    { LLAMA_LOG_INFO( "%s: CLS token        = %d '%s'\n", __func__, vocab.special_cls_id,     vocab.id_to_token[vocab.special_cls_id].text.c_str() );  }
+    if (vocab.special_mask_id != -1)    { LLAMA_LOG_INFO( "%s: MASK token       = %d '%s'\n", __func__, vocab.special_mask_id,    vocab.id_to_token[vocab.special_mask_id].text.c_str() ); }
 
-    if (vocab.linefeed_id       != -1) { LLAMA_LOG_INFO( "%s: LF token         = %d '%s'\n", __func__, vocab.linefeed_id,       vocab.id_to_token[vocab.linefeed_id].text.c_str() );       }
-    if (vocab.special_prefix_id != -1) { LLAMA_LOG_INFO( "%s: PRE token        = %d '%s'\n", __func__, vocab.special_prefix_id, vocab.id_to_token[vocab.special_prefix_id].text.c_str() ); }
-    if (vocab.special_suffix_id != -1) { LLAMA_LOG_INFO( "%s: SUF token        = %d '%s'\n", __func__, vocab.special_suffix_id, vocab.id_to_token[vocab.special_suffix_id].text.c_str() ); }
-    if (vocab.special_middle_id != -1) { LLAMA_LOG_INFO( "%s: MID token        = %d '%s'\n", __func__, vocab.special_middle_id, vocab.id_to_token[vocab.special_middle_id].text.c_str() ); }
-    if (vocab.special_eot_id    != -1) { LLAMA_LOG_INFO( "%s: EOT token        = %d '%s'\n", __func__, vocab.special_eot_id,    vocab.id_to_token[vocab.special_eot_id].text.c_str() );    }
-    if (vocab.special_eom_id    != -1) { LLAMA_LOG_INFO( "%s: EOM token        = %d '%s'\n", __func__, vocab.special_eom_id,    vocab.id_to_token[vocab.special_eom_id].text.c_str() );    }
+    if (vocab.linefeed_id != -1)        { LLAMA_LOG_INFO( "%s: LF token         = %d '%s'\n", __func__, vocab.linefeed_id,        vocab.id_to_token[vocab.linefeed_id].text.c_str() ); }
+
+    if (vocab.special_fim_pre_id != -1) { LLAMA_LOG_INFO( "%s: FIM PRE token    = %d '%s'\n", __func__, vocab.special_fim_pre_id, vocab.id_to_token[vocab.special_fim_pre_id].text.c_str() ); }
+    if (vocab.special_fim_suf_id != -1) { LLAMA_LOG_INFO( "%s: FIM SUF token    = %d '%s'\n", __func__, vocab.special_fim_suf_id, vocab.id_to_token[vocab.special_fim_suf_id].text.c_str() ); }
+    if (vocab.special_fim_mid_id != -1) { LLAMA_LOG_INFO( "%s: FIM MID token    = %d '%s'\n", __func__, vocab.special_fim_mid_id, vocab.id_to_token[vocab.special_fim_mid_id].text.c_str() ); }
+    if (vocab.special_fim_pad_id != -1) { LLAMA_LOG_INFO( "%s: FIM PAD token    = %d '%s'\n", __func__, vocab.special_fim_pad_id, vocab.id_to_token[vocab.special_fim_pad_id].text.c_str() ); }
+    if (vocab.special_fim_rep_id != -1) { LLAMA_LOG_INFO( "%s: FIM REP token    = %d '%s'\n", __func__, vocab.special_fim_rep_id, vocab.id_to_token[vocab.special_fim_rep_id].text.c_str() ); }
+    if (vocab.special_fim_sep_id != -1) { LLAMA_LOG_INFO( "%s: FIM SEP token    = %d '%s'\n", __func__, vocab.special_fim_sep_id, vocab.id_to_token[vocab.special_fim_sep_id].text.c_str() ); }
 
     for (const auto & id : vocab.special_eog_ids) {
         LLAMA_LOG_INFO( "%s: EOG token        = %d '%s'\n", __func__, id, vocab.id_to_token[id].text.c_str() );
@@ -19453,7 +19537,7 @@ struct llama_context * llama_new_context_with_model(
             }
 
             LLAMA_LOG_INFO("%s: KV self size  = %7.2f MiB, K (%s): %7.2f MiB, V (%s): %7.2f MiB\n", __func__,
-                (float)(memory_size_k + memory_size_v) / (1024.0f * 1024.0f),
+                      (float)(memory_size_k + memory_size_v) / (1024.0f * 1024.0f),
                 ggml_type_name(type_k), (float)memory_size_k / (1024.0f * 1024.0f),
                 ggml_type_name(type_v), (float)memory_size_v / (1024.0f * 1024.0f));
         }
@@ -21307,6 +21391,10 @@ llama_token llama_token_eos(const struct llama_model * model) {
     return llama_token_eos_impl(model->vocab);
 }
 
+llama_token llama_token_eot(const struct llama_model * model) {
+    return llama_token_eot_impl(model->vocab);
+}
+
 llama_token llama_token_cls(const struct llama_model * model) {
     return llama_token_cls_impl(model->vocab);
 }
@@ -21343,8 +21431,28 @@ llama_token llama_token_suffix(const struct llama_model * model) {
     return llama_token_suffix_impl(model->vocab);
 }
 
-llama_token llama_token_eot(const struct llama_model * model) {
-    return llama_token_eot_impl(model->vocab);
+llama_token llama_token_fim_pre(const struct llama_model * model) {
+    return llama_token_fim_pre_impl(model->vocab);
+}
+
+llama_token llama_token_fim_suf(const struct llama_model * model) {
+    return llama_token_fim_suf_impl(model->vocab);
+}
+
+llama_token llama_token_fim_mid(const struct llama_model * model) {
+    return llama_token_fim_mid_impl(model->vocab);
+}
+
+llama_token llama_token_fim_pad(const struct llama_model * model) {
+    return llama_token_fim_pad_impl(model->vocab);
+}
+
+llama_token llama_token_fim_rep(const struct llama_model * model) {
+    return llama_token_fim_rep_impl(model->vocab);
+}
+
+llama_token llama_token_fim_sep(const struct llama_model * model) {
+    return llama_token_fim_sep_impl(model->vocab);
 }
 
 //

From 95c76e8e92ecc93f784b185eafae36a0e7ad2fa3 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Sat, 12 Oct 2024 14:51:54 +0300
Subject: [PATCH 04/36] server : remove legacy system_prompt feature (#9857)

* server : remove legacy system_prompt feature

ggml-ci

* readme : update [no ci]

* server : fix non-transformer logic + remove response from /props
---
 common/arg.cpp             |  17 ------
 common/common.h            |   1 -
 examples/server/README.md  |   6 +--
 examples/server/server.cpp | 103 +++++++------------------------------
 4 files changed, 19 insertions(+), 108 deletions(-)

diff --git a/common/arg.cpp b/common/arg.cpp
index c4229a3a4..78cf6ab30 100644
--- a/common/arg.cpp
+++ b/common/arg.cpp
@@ -1788,23 +1788,6 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.n_threads_http = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_THREADS_HTTP"));
-    add_opt(common_arg(
-        {"-spf", "--system-prompt-file"}, "FNAME",
-        "set a file to load a system prompt (initial prompt of all slots), this is useful for chat applications",
-        [](common_params & params, const std::string & value) {
-            std::ifstream file(value);
-            if (!file) {
-                throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str()));
-            }
-            std::string system_prompt;
-            std::copy(
-                        std::istreambuf_iterator(file),
-                        std::istreambuf_iterator(),
-                        std::back_inserter(system_prompt)
-                        );
-            params.system_prompt = system_prompt;
-        }
-    ).set_examples({LLAMA_EXAMPLE_SERVER}));
     add_opt(common_arg(
         {"--metrics"},
         string_format("enable prometheus compatible metrics endpoint (default: %s)", params.endpoint_metrics ? "enabled" : "disabled"),
diff --git a/common/common.h b/common/common.h
index 5beec4bde..71e686156 100644
--- a/common/common.h
+++ b/common/common.h
@@ -282,7 +282,6 @@ struct common_params {
     std::string hostname      = "127.0.0.1";
     std::string public_path   = "";                                                                         // NOLINT
     std::string chat_template = "";                                                                         // NOLINT
-    std::string system_prompt = "";                                                                         // NOLINT
     bool enable_chat_template = true;
 
     std::vector api_keys;
diff --git a/examples/server/README.md b/examples/server/README.md
index 3da0130ac..52ccd9f5e 100644
--- a/examples/server/README.md
+++ b/examples/server/README.md
@@ -149,7 +149,6 @@ The project is under active development, and we are [looking for feedback and co
 | `--ssl-cert-file FNAME` | path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE) |
 | `-to, --timeout N` | server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT) |
 | `--threads-http N` | number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP) |
-| `-spf, --system-prompt-file FNAME` | set a file to load a system prompt (initial prompt of all slots), this is useful for chat applications |
 | `--metrics` | enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS) |
 | `--slots` | enable slots monitoring endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS) |
 | `--props` | enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS) |
@@ -320,7 +319,6 @@ node index.js
 
       - The prompt is a string or an array with the first element given as a string
       - The model's `tokenizer.ggml.add_bos_token` metadata is `true`
-      - The system prompt is empty
 
     `temperature`: Adjust the randomness of the generated text. Default: `0.8`
 
@@ -536,14 +534,12 @@ This endpoint is public (no API key check). By default, it is read-only. To make
 
 ```json
 {
-  "system_prompt": "",
   "default_generation_settings": { ... },
   "total_slots": 1,
   "chat_template": ""
 }
 ```
 
-- `system_prompt` - the system prompt (initial prompt of all slots). Please note that this does not take into account the chat template. It will append the prompt at the beginning of formatted prompt.
 - `default_generation_settings` - the default generation settings for the `/completion` endpoint, which has the same fields as the `generation_settings` response object from the `/completion` endpoint.
 - `total_slots` - the total number of slots for process requests (defined by `--parallel` option)
 - `chat_template` - the model's original Jinja2 prompt template
@@ -554,7 +550,7 @@ To use this endpoint with POST method, you need to start server with `--props`
 
 *Options:*
 
-- `system_prompt`: Change the system prompt (initial prompt of all slots). Please note that this does not take into account the chat template. It will append the prompt at the beginning of formatted prompt.
+- None yet
 
 ### POST `/v1/chat/completions`: OpenAI-compatible Chat Completions API
 
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index 314a506a1..42b57d9c4 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -623,12 +623,6 @@ struct server_context {
 
     int32_t n_ctx; // total context for all clients / slots
 
-    // system prompt
-    bool system_need_update = false;
-
-    std::string              system_prompt;
-    std::vector system_tokens;
-
     // slots / clients
     std::vector slots;
     json default_generation_settings_for_props;
@@ -665,7 +659,7 @@ struct server_context {
     bool load_model(const common_params & params_) {
         params = params_;
 
-        // dedicate one sequence to the system prompt
+        // reserve one extra sequence (seq_id == 0) for extra features
         params.n_parallel += 1;
 
         common_init_result llama_init = common_init_from_params(params);
@@ -1061,51 +1055,6 @@ struct server_context {
         clean_kv_cache = false;
     }
 
-    void system_prompt_update() {
-        SRV_DBG("updating system prompt: '%s'\n", system_prompt.c_str());
-
-        kv_cache_clear();
-        system_tokens.clear();
-
-        if (!system_prompt.empty()) {
-            system_tokens = common_tokenize(ctx, system_prompt, true);
-
-            const int32_t n_batch = llama_n_batch(ctx);
-            const int32_t n_tokens_prompt = system_tokens.size();
-
-            for (int32_t i = 0; i < n_tokens_prompt; i += n_batch) {
-                const int32_t n_tokens = std::min(n_batch, n_tokens_prompt - i);
-
-                common_batch_clear(batch);
-
-                for (int32_t j = 0; j < n_tokens; ++j) {
-                    common_batch_add(batch, system_tokens[i + j], i + j, { 0 }, false);
-                }
-
-                if (llama_decode(ctx, batch) != 0) {
-                    SRV_ERR("%s", "llama_decode() failed\n");
-                    return;
-                }
-            }
-
-            // assign the system KV cache to all parallel sequences
-            for (int32_t i = 1; i <= params.n_parallel; ++i) {
-                llama_kv_cache_seq_cp(ctx, 0, i, -1, -1);
-            }
-        }
-
-        system_need_update = false;
-    }
-
-    bool system_prompt_set(const std::string & sys_prompt) {
-        SRV_DBG("system prompt set: '%s'\n", system_prompt.c_str());
-
-        system_prompt = sys_prompt;
-        // update system_tokens and KV cache as soon as all slots are idle
-        system_need_update = true;
-        return true;
-    }
-
     bool process_token(completion_token_output & result, server_slot & slot) {
         // remember which tokens were sampled - used for repetition penalties during sampling
         const std::string token_str = common_token_to_piece(ctx, result.tok, params.special);
@@ -1855,12 +1804,8 @@ struct server_context {
             }
 
             if (all_idle) {
-                if (system_need_update) {
-                    system_prompt_update();
-                }
-
                 SRV_INF("%s", "all slots are idle\n");
-                if (system_prompt.empty() && clean_kv_cache) {
+                if (clean_kv_cache) {
                     kv_cache_clear();
                 }
 
@@ -1882,7 +1827,7 @@ struct server_context {
         // TODO: simplify and improve
         for (server_slot & slot : slots) {
             if (slot.ga_n == 1) {
-                if (slot.is_processing() && (int) system_tokens.size() + slot.n_past >= slot.n_ctx - 1) {
+                if (slot.is_processing() && slot.n_past >= slot.n_ctx - 1) {
                     if (!params.ctx_shift) {
                         // this check is redundant (for good)
                         // we should never get here, because generation should already stopped in process_token()
@@ -1893,13 +1838,13 @@ struct server_context {
 
                     // Shift context
                     const int n_keep    = slot.params.n_keep + add_bos_token;
-                    const int n_left    = (int) system_tokens.size() + slot.n_past - n_keep;
+                    const int n_left    = slot.n_past - n_keep;
                     const int n_discard = slot.params.n_discard ? slot.params.n_discard : (n_left / 2);
 
                     SLT_WRN(slot, "slot context shift, n_keep = %d, n_left = %d, n_discard = %d\n", n_keep, n_left, n_discard);
 
                     llama_kv_cache_seq_rm (ctx, slot.id + 1, n_keep            , n_keep + n_discard);
-                    llama_kv_cache_seq_add(ctx, slot.id + 1, n_keep + n_discard, system_tokens.size() + slot.n_past, -n_discard);
+                    llama_kv_cache_seq_add(ctx, slot.id + 1, n_keep + n_discard, slot.n_past,        -n_discard);
 
                     if (slot.params.cache_prompt) {
                         for (size_t i = n_keep + n_discard; i < slot.cache_tokens.size(); i++) {
@@ -1929,9 +1874,7 @@ struct server_context {
 
             const int32_t slot_npast = slot.n_past_se > 0 ? slot.n_past_se : slot.n_past;
 
-            // TODO: we always have to take into account the "system_tokens"
-            //       this is not great and needs to be improved somehow
-            common_batch_add(batch, slot.sampled, system_tokens.size() + slot_npast, { slot.id + 1 }, true);
+            common_batch_add(batch, slot.sampled, slot_npast, { slot.id + 1 }, true);
 
             slot.n_past += 1;
 
@@ -1939,8 +1882,8 @@ struct server_context {
                 slot.cache_tokens.push_back(slot.sampled);
             }
 
-            SLT_DBG(slot, "slot decode token, n_ctx = %d, n_past = %d, n_system_tokens = %d, n_cache_tokens = %d, truncated = %d\n",
-                    slot.n_ctx, slot.n_past, (int) system_tokens.size(), (int) slot.cache_tokens.size(), slot.truncated);
+            SLT_DBG(slot, "slot decode token, n_ctx = %d, n_past = %d, n_cache_tokens = %d, truncated = %d\n",
+                    slot.n_ctx, slot.n_past, (int) slot.cache_tokens.size(), slot.truncated);
         }
 
         // process in chunks of params.n_batch
@@ -1971,7 +1914,7 @@ struct server_context {
                             case SERVER_TASK_CMPL_TYPE_NORMAL:
                             case SERVER_TASK_CMPL_TYPE_EMBEDDING:
                                 {
-                                    prompt_tokens = tokenize(slot.prompt, system_prompt.empty(), true); // add BOS if there isn't system prompt
+                                    prompt_tokens = tokenize(slot.prompt, llama_add_bos_token(model), true);
                                 } break;
                             case SERVER_TASK_CMPL_TYPE_RERANK:
                                 {
@@ -2050,7 +1993,7 @@ struct server_context {
                         } else {
                             if (!params.ctx_shift) {
                                 // if context shift is disabled, we make sure prompt size is smaller than KV size
-                                if ((int) system_tokens.size() + slot.n_prompt_tokens >= slot.n_ctx) {
+                                if (slot.n_prompt_tokens >= slot.n_ctx) {
                                     slot.release();
                                     send_error(slot, "the request exceeds the available context size. try increasing the context size or enable context shift", ERROR_TYPE_INVALID_REQUEST);
                                     continue;
@@ -2138,22 +2081,19 @@ struct server_context {
                     }
 
                     // keep only the common part
-                    int p0 = (int) system_tokens.size() + slot.n_past;
+                    int p0 = slot.n_past;
+
                     if (!llama_kv_cache_seq_rm(ctx, slot.id + 1, p0, -1)) {
                         // could not partially delete (likely using a non-Transformer model)
                         llama_kv_cache_seq_rm(ctx, slot.id + 1, -1, -1);
 
-                        p0 = (int) system_tokens.size();
-                        if (p0 != 0) {
-                            // copy over the system prompt when there is one
-                            llama_kv_cache_seq_cp(ctx, 0, slot.id + 1, -1, -1);
-                        }
+                        p0 = 0;
 
-                        // there is no common part left (except for the system prompt)
+                        // there is no common part left
                         slot.n_past = 0;
                         slot.n_past_se = 0;
                         slot.ga_i = 0;
-                        // TODO: is the system prompt ever in the sampling context?
+
                         common_sampler_reset(slot.smpl);
                     }
 
@@ -2179,7 +2119,7 @@ struct server_context {
                             }
                         }
 
-                        common_batch_add(batch, prompt_tokens[slot.n_past], system_tokens.size() + slot_npast, { slot.id + 1 }, false);
+                        common_batch_add(batch, prompt_tokens[slot.n_past], slot_npast, { slot.id + 1 }, false);
 
                         if (slot.params.cache_prompt) {
                             slot.cache_tokens.push_back(prompt_tokens[slot.n_past]);
@@ -2409,10 +2349,6 @@ int main(int argc, char ** argv) {
     // struct that contains llama context and inference
     server_context ctx_server;
 
-    if (!params.system_prompt.empty()) {
-        ctx_server.system_prompt_set(params.system_prompt);
-    }
-
     if (params.model_alias == "unknown") {
         params.model_alias = params.model;
     }
@@ -2840,7 +2776,6 @@ int main(int argc, char ** argv) {
 
     const auto handle_props = [&ctx_server, &res_ok](const httplib::Request &, httplib::Response & res) {
         json data = {
-            { "system_prompt",               ctx_server.system_prompt },
             { "default_generation_settings", ctx_server.default_generation_settings_for_props },
             { "total_slots",                 ctx_server.params.n_parallel },
             { "chat_template",               llama_get_chat_template(ctx_server.model) },
@@ -2856,10 +2791,8 @@ int main(int argc, char ** argv) {
         }
 
         json data = json::parse(req.body);
-        if (data.contains("system_prompt")) {
-            std::string system_prompt = data.at("system_prompt");
-            ctx_server.system_prompt_set(system_prompt);
-        }
+
+        // update any props here
 
         res_ok(res, {{ "success", true }});
     };

From 1bde94dd024b632f98428f4bf2ce483295130779 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Sat, 12 Oct 2024 16:06:31 +0300
Subject: [PATCH 05/36] server : remove self-extend features (#9860)

* server : remove self-extend

ggml-ci

* server : fix context limit check to use slot.n_past

ggml-ci
---
 common/arg.cpp                                |   6 +-
 examples/server/README.md                     |   2 -
 examples/server/server.cpp                    | 187 +++++-------------
 .../server/tests/features/ctx_shift.feature   |   4 +
 4 files changed, 57 insertions(+), 142 deletions(-)

diff --git a/common/arg.cpp b/common/arg.cpp
index 78cf6ab30..205177d46 100644
--- a/common/arg.cpp
+++ b/common/arg.cpp
@@ -1163,14 +1163,14 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         [](common_params & params, int value) {
             params.grp_attn_n = value;
         }
-    ).set_env("LLAMA_ARG_GRP_ATTN_N"));
+    ).set_env("LLAMA_ARG_GRP_ATTN_N").set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_PASSKEY}));
     add_opt(common_arg(
         {"-gaw", "--grp-attn-w"}, "N",
-        string_format("group-attention width (default: %.1f)", (double)params.grp_attn_w),
+        string_format("group-attention width (default: %d)", params.grp_attn_w),
         [](common_params & params, int value) {
             params.grp_attn_w = value;
         }
-    ).set_env("LLAMA_ARG_GRP_ATTN_W"));
+    ).set_env("LLAMA_ARG_GRP_ATTN_W").set_examples({LLAMA_EXAMPLE_MAIN}));
     add_opt(common_arg(
         {"-dkvc", "--dump-kv-cache"},
         "verbose print of the KV cache",
diff --git a/examples/server/README.md b/examples/server/README.md
index 52ccd9f5e..caffbac52 100644
--- a/examples/server/README.md
+++ b/examples/server/README.md
@@ -60,8 +60,6 @@ The project is under active development, and we are [looking for feedback and co
 | `--yarn-attn-factor N` | YaRN: scale sqrt(t) or attention magnitude (default: 1.0)
(env: LLAMA_ARG_YARN_ATTN_FACTOR) |
 | `--yarn-beta-slow N` | YaRN: high correction dim or alpha (default: 1.0)
(env: LLAMA_ARG_YARN_BETA_SLOW) |
 | `--yarn-beta-fast N` | YaRN: low correction dim or beta (default: 32.0)
(env: LLAMA_ARG_YARN_BETA_FAST) |
-| `-gan, --grp-attn-n N` | group-attention factor (default: 1)
(env: LLAMA_ARG_GRP_ATTN_N) |
-| `-gaw, --grp-attn-w N` | group-attention width (default: 512.0)
(env: LLAMA_ARG_GRP_ATTN_W) |
 | `-dkvc, --dump-kv-cache` | verbose print of the KV cache |
 | `-nkvo, --no-kv-offload` | disable KV offload
(env: LLAMA_ARG_NO_KV_OFFLOAD) |
 | `-ctk, --cache-type-k TYPE` | KV cache data type for K (default: f16)
(env: LLAMA_ARG_CACHE_TYPE_K) |
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index 42b57d9c4..0dd2fc8b2 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -193,21 +193,15 @@ struct server_slot {
 
     llama_token sampled;
 
-    int32_t ga_i = 0;   // group-attention state
-    int32_t ga_n = 1;   // group-attention factor
-    int32_t ga_w = 512; // group-attention width
-
-    int32_t n_past_se = 0; // self-extend
-
     // stats
-    size_t n_sent_text = 0; // number of sent text character
+    size_t n_sent_text        = 0; // number of sent text character
     size_t n_sent_token_probs = 0;
 
     int64_t t_start_process_prompt;
     int64_t t_start_generation;
 
     double t_prompt_processing; // ms
-    double t_token_generation; // ms
+    double t_token_generation;  // ms
 
     std::function callback_on_release;
 
@@ -225,8 +219,6 @@ struct server_slot {
         n_sent_text        = 0;
         n_sent_token_probs = 0;
         cmpl_type          = SERVER_TASK_CMPL_TYPE_NORMAL;
-        ga_i               = 0;
-        n_past_se          = 0;
 
         generated_token_probs.clear();
     }
@@ -705,22 +697,6 @@ struct server_context {
 
             SLT_INF(slot, "new slot n_ctx_slot = %d\n", slot.n_ctx);
 
-            const int ga_n = params.grp_attn_n;
-            const int ga_w = params.grp_attn_w;
-
-            if (ga_n != 1) {
-                GGML_ASSERT(ga_n > 0                    && "ga_n must be positive");                       // NOLINT
-                GGML_ASSERT(ga_w % ga_n == 0            && "ga_w must be a multiple of ga_n");             // NOLINT
-                //GGML_ASSERT(n_ctx_train % ga_w == 0     && "n_ctx_train must be a multiple of ga_w");    // NOLINT
-                //GGML_ASSERT(n_ctx >= n_ctx_train * ga_n && "n_ctx must be at least n_ctx_train * ga_n"); // NOLINT
-
-                SLT_INF(slot, "slot self-extend: ga_n = %d, ga_w = %d\n", ga_n, ga_w);
-            }
-
-            slot.ga_i = 0;
-            slot.ga_n = ga_n;
-            slot.ga_w = ga_w;
-
             slot.sparams = params.sparams;
 
             slot.callback_on_release = [this](int) {
@@ -906,19 +882,14 @@ struct server_context {
         }
         if (data.contains("json_schema") && !data.contains("grammar")) {
             try {
-                auto schema                = json_value(data, "json_schema", json::object());
-                slot.sparams.grammar       = json_schema_to_grammar(schema);
+                auto schema          = json_value(data, "json_schema", json::object());
+                slot.sparams.grammar = json_schema_to_grammar(schema);
             } catch (const std::exception & e) {
                 send_error(task, std::string("\"json_schema\": ") + e.what(), ERROR_TYPE_INVALID_REQUEST);
                 return false;
             }
         } else {
-            slot.sparams.grammar       = json_value(data, "grammar",           default_sparams.grammar);
-        }
-
-        if (slot.params.cache_prompt && slot.ga_n != 1) {
-            slot.params.cache_prompt = false;
-            SLT_WRN(slot, "%s", "group-attention is not supported with prompt caching. disabling cache\n");
+            slot.sparams.grammar = json_value(data, "grammar", default_sparams.grammar);
         }
 
         if (slot.n_predict > 0 && slot.params.n_predict > slot.n_predict) {
@@ -1131,12 +1102,13 @@ struct server_context {
         }
 
         // if context shift is disabled, we stop when it reaches the context limit
-        if (slot.n_decoded >= slot.n_ctx) {
+        if (slot.n_past >= slot.n_ctx) {
             slot.truncated      = true;
             slot.stopped_limit  = true;
             slot.has_next_token = false;
 
-            SLT_DBG(slot, "stopped due to running out of context capacity, n_decoded = %d, n_ctx = %d\n", slot.n_decoded, slot.n_ctx);
+            SLT_DBG(slot, "stopped due to running out of context capacity, n_past = %d, n_prompt_tokens = %d, n_decoded = %d, n_ctx = %d\n",
+                    slot.n_decoded, slot.n_prompt_tokens, slot.n_past, slot.n_ctx);
         }
 
         if (llama_token_is_eog(model, result.tok)) {
@@ -1148,13 +1120,13 @@ struct server_context {
 
         const auto n_ctx_train = llama_n_ctx_train(model);
 
-        if (slot.params.n_predict < 1 && slot.n_predict < 1 && slot.ga_n == 1 && slot.n_prompt_tokens + slot.n_decoded >= n_ctx_train) {
+        if (slot.params.n_predict < 1 && slot.n_predict < 1 && slot.n_prompt_tokens + slot.n_decoded >= n_ctx_train) {
             slot.truncated      = true;
             slot.stopped_limit  = true;
             slot.has_next_token = false; // stop prediction
 
             SLT_WRN(slot,
-                    "n_predict (%d) is not set and self-context extend is disabled. "
+                    "n_predict (%d) is set for infinite generation. "
                     "Limiting generated tokens to n_ctx_train (%d) to avoid EOS-less generation infinite loop\n",
                     slot.params.n_predict, n_ctx_train);
         }
@@ -1826,38 +1798,36 @@ struct server_context {
         // apply context-shift if needed
         // TODO: simplify and improve
         for (server_slot & slot : slots) {
-            if (slot.ga_n == 1) {
-                if (slot.is_processing() && slot.n_past >= slot.n_ctx - 1) {
-                    if (!params.ctx_shift) {
-                        // this check is redundant (for good)
-                        // we should never get here, because generation should already stopped in process_token()
-                        slot.release();
-                        send_error(slot, "context shift is disabled", ERROR_TYPE_SERVER);
-                        continue;
-                    }
-
-                    // Shift context
-                    const int n_keep    = slot.params.n_keep + add_bos_token;
-                    const int n_left    = slot.n_past - n_keep;
-                    const int n_discard = slot.params.n_discard ? slot.params.n_discard : (n_left / 2);
-
-                    SLT_WRN(slot, "slot context shift, n_keep = %d, n_left = %d, n_discard = %d\n", n_keep, n_left, n_discard);
-
-                    llama_kv_cache_seq_rm (ctx, slot.id + 1, n_keep            , n_keep + n_discard);
-                    llama_kv_cache_seq_add(ctx, slot.id + 1, n_keep + n_discard, slot.n_past,        -n_discard);
-
-                    if (slot.params.cache_prompt) {
-                        for (size_t i = n_keep + n_discard; i < slot.cache_tokens.size(); i++) {
-                            slot.cache_tokens[i - n_discard] = slot.cache_tokens[i];
-                        }
-
-                        slot.cache_tokens.resize(slot.cache_tokens.size() - n_discard);
-                    }
-
-                    slot.n_past -= n_discard;
-
-                    slot.truncated = true;
+            if (slot.is_processing() && slot.n_past + 1 >= slot.n_ctx) {
+                if (!params.ctx_shift) {
+                    // this check is redundant (for good)
+                    // we should never get here, because generation should already stopped in process_token()
+                    slot.release();
+                    send_error(slot, "context shift is disabled", ERROR_TYPE_SERVER);
+                    continue;
                 }
+
+                // Shift context
+                const int n_keep    = slot.params.n_keep + add_bos_token;
+                const int n_left    = slot.n_past - n_keep;
+                const int n_discard = slot.params.n_discard ? slot.params.n_discard : (n_left / 2);
+
+                SLT_WRN(slot, "slot context shift, n_keep = %d, n_left = %d, n_discard = %d\n", n_keep, n_left, n_discard);
+
+                llama_kv_cache_seq_rm (ctx, slot.id + 1, n_keep            , n_keep + n_discard);
+                llama_kv_cache_seq_add(ctx, slot.id + 1, n_keep + n_discard, slot.n_past,        -n_discard);
+
+                if (slot.params.cache_prompt) {
+                    for (size_t i = n_keep + n_discard; i < slot.cache_tokens.size(); i++) {
+                        slot.cache_tokens[i - n_discard] = slot.cache_tokens[i];
+                    }
+
+                    slot.cache_tokens.resize(slot.cache_tokens.size() - n_discard);
+                }
+
+                slot.n_past -= n_discard;
+
+                slot.truncated = true;
             }
         }
 
@@ -1872,9 +1842,7 @@ struct server_context {
 
             slot.i_batch = batch.n_tokens;
 
-            const int32_t slot_npast = slot.n_past_se > 0 ? slot.n_past_se : slot.n_past;
-
-            common_batch_add(batch, slot.sampled, slot_npast, { slot.id + 1 }, true);
+            common_batch_add(batch, slot.sampled, slot.n_past, { slot.id + 1 }, true);
 
             slot.n_past += 1;
 
@@ -1993,6 +1961,8 @@ struct server_context {
                         } else {
                             if (!params.ctx_shift) {
                                 // if context shift is disabled, we make sure prompt size is smaller than KV size
+                                // TODO: there should be a separate parameter that control prompt truncation
+                                //       context shift should be applied only during the generation phase
                                 if (slot.n_prompt_tokens >= slot.n_ctx) {
                                     slot.release();
                                     send_error(slot, "the request exceeds the available context size. try increasing the context size or enable context shift", ERROR_TYPE_INVALID_REQUEST);
@@ -2005,7 +1975,7 @@ struct server_context {
                             slot.params.n_keep = std::min(slot.n_ctx - 4, slot.params.n_keep);
 
                             // if input prompt is too big, truncate it (if group attention self-extend is disabled)
-                            if (slot.ga_n == 1 && slot.n_prompt_tokens >= slot.n_ctx) {
+                            if (slot.n_prompt_tokens >= slot.n_ctx) {
                                 const int n_left = slot.n_ctx - slot.params.n_keep;
 
                                 const int n_block_size = n_left / 2;
@@ -2032,12 +2002,7 @@ struct server_context {
 
                             common_sampler_reset(slot.smpl);
 
-                            if (!slot.params.cache_prompt) {
-                                slot.n_past_se = 0;
-                                slot.ga_i      = 0;
-                            } else {
-                                GGML_ASSERT(slot.ga_n == 1);
-
+                            if (slot.params.cache_prompt) {
                                 // reuse any previously computed tokens that are common with the new prompt
                                 slot.n_past = common_part(slot.cache_tokens, prompt_tokens);
 
@@ -2053,9 +2018,6 @@ struct server_context {
                             SLT_WRN(slot, "need to evaluate at least 1 token to generate logits, n_past = %d, n_prompt_tokens = %d\n", slot.n_past, slot.n_prompt_tokens);
 
                             slot.n_past--;
-                            if (slot.ga_i > 0) {
-                                slot.n_past_se--;
-                            }
                         }
 
                         slot.n_prompt_tokens_processed = 0;
@@ -2081,52 +2043,31 @@ struct server_context {
                     }
 
                     // keep only the common part
-                    int p0 = slot.n_past;
-
-                    if (!llama_kv_cache_seq_rm(ctx, slot.id + 1, p0, -1)) {
+                    if (!llama_kv_cache_seq_rm(ctx, slot.id + 1, slot.n_past, -1)) {
                         // could not partially delete (likely using a non-Transformer model)
                         llama_kv_cache_seq_rm(ctx, slot.id + 1, -1, -1);
 
-                        p0 = 0;
-
                         // there is no common part left
                         slot.n_past = 0;
-                        slot.n_past_se = 0;
-                        slot.ga_i = 0;
 
                         common_sampler_reset(slot.smpl);
                     }
 
+                    SLT_INF(slot, "kv cache rm [%d, end)\n", slot.n_past);
+
                     // remove the non-common part from the cache
                     slot.cache_tokens.resize(slot.n_past);
 
-                    SLT_INF(slot, "kv cache rm [%d, end)\n", p0);
-
-                    int32_t slot_npast = slot.n_past_se > 0 ? slot.n_past_se : slot.n_past;
-
-                    int32_t ga_i = slot.ga_i;
-                    int32_t ga_n = slot.ga_n;
-                    int32_t ga_w = slot.ga_w;
-
                     // add prompt tokens for processing in the current batch
-                    // TODO: the self-extend stuff here is a mess - simplify and/or abstract it somehow
-                    for (; slot.n_past < slot.n_prompt_tokens && batch.n_tokens < n_batch; ++slot.n_past) {
-                        if (slot.ga_n != 1) {
-                            while (slot_npast >= ga_i + ga_w) {
-                                const int bd = (ga_w/ga_n)*(ga_n - 1);
-                                slot_npast -= bd;
-                                ga_i += ga_w/ga_n;
-                            }
-                        }
-
-                        common_batch_add(batch, prompt_tokens[slot.n_past], slot_npast, { slot.id + 1 }, false);
+                    while (slot.n_past < slot.n_prompt_tokens && batch.n_tokens < n_batch) {
+                        common_batch_add(batch, prompt_tokens[slot.n_past], slot.n_past, { slot.id + 1 }, false);
 
                         if (slot.params.cache_prompt) {
                             slot.cache_tokens.push_back(prompt_tokens[slot.n_past]);
                         }
 
                         slot.n_prompt_tokens_processed++;
-                        slot_npast++;
+                        slot.n_past++;
                     }
 
                     SLT_INF(slot, "prompt processing progress, n_past = %d, n_tokens = %d, progress = %f\n", slot.n_past, batch.n_tokens, (float) slot.n_prompt_tokens_processed / slot.n_prompt_tokens);
@@ -2167,34 +2108,6 @@ struct server_context {
         for (int32_t i = 0; i < batch.n_tokens; i += n_batch) {
             const int32_t n_tokens = std::min(n_batch, batch.n_tokens - i);
 
-            for (auto & slot : slots) {
-                if (slot.ga_n != 1) {
-                    // context extension via Self-Extend
-                    // TODO: simplify and/or abstract this
-                    while (slot.n_past_se >= slot.ga_i + slot.ga_w) {
-                        const int ib = (slot.ga_n * slot.ga_i) / slot.ga_w;
-                        const int bd = (slot.ga_w / slot.ga_n) * (slot.ga_n - 1);
-                        const int dd = (slot.ga_w / slot.ga_n) - ib * bd - slot.ga_w;
-
-                        SLT_DBG(slot, "shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i, slot.n_past_se, ib * bd, slot.ga_i + ib * bd, slot.n_past_se + ib * bd);
-                        SLT_DBG(slot, "div:   [%6d, %6d] / %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w, slot.ga_n, (slot.ga_i + ib * bd) / slot.ga_n, (slot.ga_i + ib * bd + slot.ga_w) / slot.ga_n);
-                        SLT_DBG(slot, "shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", slot.ga_i + ib * bd + slot.ga_w, slot.n_past_se + ib * bd, dd, slot.ga_i + ib * bd + slot.ga_w + dd, slot.n_past_se + ib * bd + dd);
-
-                        llama_kv_cache_seq_add(ctx, slot.id + 1, slot.ga_i, slot.n_past_se, ib * bd);
-                        llama_kv_cache_seq_div(ctx, slot.id + 1, slot.ga_i + ib * bd, slot.ga_i + ib * bd + slot.ga_w, slot.ga_n);
-                        llama_kv_cache_seq_add(ctx, slot.id + 1, slot.ga_i + ib * bd + slot.ga_w, slot.n_past_se + ib * bd, dd);
-
-                        slot.n_past_se -= bd;
-
-                        slot.ga_i += slot.ga_w / slot.ga_n;
-
-                        SLT_DBG(slot, "\nn_past_old = %d, n_past = %d, ga_i = %d\n\n", slot.n_past_se + bd, slot.n_past_se, slot.ga_i);
-                    }
-
-                    slot.n_past_se += n_tokens;
-                }
-            }
-
             llama_batch batch_view = {
                 n_tokens,
                 batch.token    + i,
diff --git a/examples/server/tests/features/ctx_shift.feature b/examples/server/tests/features/ctx_shift.feature
index ba3afcf06..ae6c6b01b 100644
--- a/examples/server/tests/features/ctx_shift.feature
+++ b/examples/server/tests/features/ctx_shift.feature
@@ -13,6 +13,10 @@ Feature: llama.cpp server
     And   32 as batch size
     And   2 slots
 
+    # the prompt is 301 tokens
+    # the slot context is 256/2 = 128 tokens
+    # the prompt is truncated to keep the last 109 tokens
+    # 64 tokens are generated thanks to shifting the context when it gets full
   Scenario: Inference with context shift
     And   64 server max tokens to predict
     Then  the server is starting

From edc265661cd707327297b6ec4d83423c43cb50a5 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Sat, 12 Oct 2024 16:14:27 +0300
Subject: [PATCH 06/36] server : add option to time limit the generation phase
 (#9865)

ggml-ci
---
 examples/server/README.md  |  2 ++
 examples/server/server.cpp | 50 +++++++++++++++++++++++++++++++++-----
 2 files changed, 46 insertions(+), 6 deletions(-)

diff --git a/examples/server/README.md b/examples/server/README.md
index caffbac52..b5feeb77b 100644
--- a/examples/server/README.md
+++ b/examples/server/README.md
@@ -374,6 +374,8 @@ node index.js
 
     `min_keep`: If greater than 0, force samplers to return N possible tokens at minimum. Default: `0`
 
+    `t_max_predict_ms`: Set a time limit in milliseconds for the prediction (a.k.a. text-generation) phase. The timeout will trigger if the generation takes more than the specified time (measured since the first token was generated) and if a new-line character has already been generated. Useful for FIM applications. Default: `0`, which is disabled.
+
     `image_data`: An array of objects to hold base64-encoded image `data` and its `id`s to be reference in `prompt`. You can determine the place of the image in the prompt as in the following: `USER:[img-12]Describe the image in detail.\nASSISTANT:`. In this case, `[img-12]` will be replaced by the embeddings of the image with id `12` in the following `image_data` array: `{..., "image_data": [{"data": "", "id": 12}]}`. Use `image_data` only with multimodal models, e.g., LLaVA.
 
     `id_slot`: Assign the completion task to an specific slot. If is -1 the task will be assigned to a Idle slot.  Default: `-1`
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index 0dd2fc8b2..f809c46d5 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -128,9 +128,12 @@ struct slot_params {
     bool stream       = true;
     bool cache_prompt = false; // remember the prompt to avoid reprocessing all prompt
 
-    int32_t  n_keep    =  0; // number of tokens to keep from initial prompt
-    int32_t  n_discard =  0; // number of tokens after n_keep that may be discarded when shifting context, 0 defaults to half
-    int32_t  n_predict = -1; // new tokens to predict
+    int32_t n_keep    =  0; // number of tokens to keep from initial prompt
+    int32_t n_discard =  0; // number of tokens after n_keep that may be discarded when shifting context, 0 defaults to half
+    int32_t n_predict = -1; // new tokens to predict
+
+    int64_t t_max_prompt_ms  = -1; // TODO: implement
+    int64_t t_max_predict_ms = -1; // if positive, limit the generation phase to this time limit
 
     std::vector antiprompt;
 
@@ -175,6 +178,7 @@ struct server_slot {
     server_task_cmpl_type cmpl_type = SERVER_TASK_CMPL_TYPE_NORMAL;
 
     bool has_next_token = true;
+    bool has_new_line   = false;
     bool truncated      = false;
     bool stopped_eos    = false;
     bool stopped_word   = false;
@@ -210,6 +214,7 @@ struct server_slot {
 
         n_prompt_tokens    = 0;
         generated_text     = "";
+        has_new_line       = false;
         truncated          = false;
         stopped_eos        = false;
         stopped_word       = false;
@@ -874,6 +879,8 @@ struct server_context {
         slot.sparams.seed              = json_value(data, "seed",              default_sparams.seed);
         slot.sparams.n_probs           = json_value(data, "n_probs",           default_sparams.n_probs);
         slot.sparams.min_keep          = json_value(data, "min_keep",          default_sparams.min_keep);
+      //slot.params.t_max_prompt_ms    = json_value(data, "t_max_prompt_ms",   default_params.t_max_prompt_ms); // TODO: implement
+        slot.params.t_max_predict_ms   = json_value(data, "t_max_predict_ms",  default_params.t_max_predict_ms);
 
         // process "json_schema" and "grammar"
         if (data.contains("json_schema") && !data.at("json_schema").is_null() && data.contains("grammar") && !data.at("grammar").is_null()) {
@@ -1101,6 +1108,20 @@ struct server_context {
             SLT_DBG(slot, "stopped by limit, n_decoded = %d, n_predict = %d\n", slot.n_decoded, slot.params.n_predict);
         }
 
+        // if we have already seen a new line, we stop after a certain time limit
+        if (slot.has_new_line && slot.params.t_max_predict_ms > 0 &&
+            (ggml_time_us() - slot.t_start_generation > 1000.0f*slot.params.t_max_predict_ms)) {
+            slot.stopped_limit  = true;
+            slot.has_next_token = false;
+
+            SLT_DBG(slot, "stopped by time limit, n_decoded = %d, t_max_predict_ms = %d ms\n", slot.n_decoded, (int) slot.params.t_max_predict_ms);
+        }
+
+        // check if there is a new line in the generated text
+        if (result.text_to_send.find('\n') != std::string::npos) {
+            slot.has_new_line = true;
+        }
+
         // if context shift is disabled, we stop when it reaches the context limit
         if (slot.n_past >= slot.n_ctx) {
             slot.truncated      = true;
@@ -1250,6 +1271,7 @@ struct server_context {
             {"tokens_evaluated",    slot.n_prompt_tokens},
             {"generation_settings", get_formated_generation(slot)},
             {"prompt",              slot.prompt},
+            {"has_new_line",        slot.has_new_line},
             {"truncated",           slot.truncated},
             {"stopped_eos",         slot.stopped_eos},
             {"stopped_word",        slot.stopped_word},
@@ -1576,6 +1598,7 @@ struct server_context {
                         slot_data["prompt"]     = slot.prompt;
                         slot_data["next_token"] = {
                             {"has_next_token", slot.has_next_token},
+                            {"has_new_line",   slot.has_new_line},
                             {"n_remain",       slot.n_remaining},
                             {"n_decoded",      slot.n_decoded},
                             {"stopped_eos",    slot.stopped_eos},
@@ -1914,6 +1937,13 @@ struct server_context {
                                     auto prefix_tokens = tokenize(slot.params.input_prefix, false, false);
                                     auto suffix_tokens = tokenize(slot.params.input_suffix, false, false);
 
+                                    // for now pick context to fit in a single batch (ratio prefix:suffix = 3:1, TODO: configurable?)
+                                    const int n_suffix_take = std::min(suffix_tokens.size(), n_batch/4);
+                                    const int n_prefix_take = std::min(prefix_tokens.size(), (n_batch - 3) - n_suffix_take);
+
+                                    prefix_tokens.erase(prefix_tokens.begin(), prefix_tokens.begin() + prefix_tokens.size() - n_prefix_take);
+                                    suffix_tokens.resize(n_suffix_take);
+
                                     prefix_tokens.insert(prefix_tokens.begin(), llama_token_fim_pre(model));
                                     suffix_tokens.insert(suffix_tokens.begin(), llama_token_fim_suf(model));
 
@@ -1936,9 +1966,17 @@ struct server_context {
 
                         SLT_INF(slot, "prompt tokenized, n_ctx_slot = %d, n_keep = %d, n_prompt_tokens = %d\n", slot.n_ctx, slot.params.n_keep, slot.n_prompt_tokens);
 
-                        // print prompt tokens:
-                        for (int i = 0; i < (int) prompt_tokens.size(); i++) {
-                            SLT_DBG(slot, "prompt token %3d: %6d '%s'\n", i, prompt_tokens[i], common_token_to_piece(ctx, prompt_tokens[i]).c_str());
+                        // print prompt tokens (for debugging)
+                        if (1) {
+                            // first 16 tokens (avoid flooding logs)
+                            for (int i = 0; i < std::min(16, prompt_tokens.size()); i++) {
+                                SLT_DBG(slot, "prompt token %3d: %6d '%s'\n", i, prompt_tokens[i], common_token_to_piece(ctx, prompt_tokens[i]).c_str());
+                            }
+                        } else {
+                            // all
+                            for (int i = 0; i < (int) prompt_tokens.size(); i++) {
+                                SLT_DBG(slot, "prompt token %3d: %6d '%s'\n", i, prompt_tokens[i], common_token_to_piece(ctx, prompt_tokens[i]).c_str());
+                            }
                         }
 
                         // empty prompt passed -> release the slot and send empty response

From 92be9f12164f18ce845a5bab60cefa5f7fec6836 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Sun, 13 Oct 2024 06:11:26 +0300
Subject: [PATCH 07/36] flake.lock: Update (#9870)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Flake lock file updates:

• Updated input 'nixpkgs':
    'github:NixOS/nixpkgs/bc947f541ae55e999ffdb4013441347d83b00feb?narHash=sha256-NOiTvBbRLIOe5F6RbHaAh6%2B%2BBNjsb149fGZd1T4%2BKBg%3D' (2024-10-04)
  → 'github:NixOS/nixpkgs/5633bcff0c6162b9e4b5f1264264611e950c8ec7?narHash=sha256-9UTxR8eukdg%2BXZeHgxW5hQA9fIKHsKCdOIUycTryeVw%3D' (2024-10-09)

Co-authored-by: github-actions[bot] 
---
 flake.lock | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/flake.lock b/flake.lock
index 3fb6ced51..702527028 100644
--- a/flake.lock
+++ b/flake.lock
@@ -20,11 +20,11 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1728018373,
-        "narHash": "sha256-NOiTvBbRLIOe5F6RbHaAh6++BNjsb149fGZd1T4+KBg=",
+        "lastModified": 1728492678,
+        "narHash": "sha256-9UTxR8eukdg+XZeHgxW5hQA9fIKHsKCdOIUycTryeVw=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "bc947f541ae55e999ffdb4013441347d83b00feb",
+        "rev": "5633bcff0c6162b9e4b5f1264264611e950c8ec7",
         "type": "github"
       },
       "original": {

From c7181bd294757dd80a7904e3dd0fea2d0be914e7 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Sun, 13 Oct 2024 18:52:48 +0300
Subject: [PATCH 08/36] server : reuse cached context chunks (#9866)

ggml-ci
---
 common/arg.cpp             |  7 ++++
 common/common.h            |  3 +-
 examples/server/README.md  |  1 +
 examples/server/server.cpp | 69 ++++++++++++++++++++++++++++++++++++--
 examples/server/utils.hpp  |  4 +--
 5 files changed, 78 insertions(+), 6 deletions(-)

diff --git a/common/arg.cpp b/common/arg.cpp
index 205177d46..8969fc107 100644
--- a/common/arg.cpp
+++ b/common/arg.cpp
@@ -1788,6 +1788,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.n_threads_http = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_THREADS_HTTP"));
+    add_opt(common_arg(
+        {"--cache-reuse"}, "N",
+        string_format("min chunk size to attempt reusing from the cache via KV shifting (default: %d)", params.n_cache_reuse),
+        [](common_params & params, int value) {
+            params.n_cache_reuse = value;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CACHE_REUSE"));
     add_opt(common_arg(
         {"--metrics"},
         string_format("enable prometheus compatible metrics endpoint (default: %s)", params.endpoint_metrics ? "enabled" : "disabled"),
diff --git a/common/common.h b/common/common.h
index 71e686156..5507b1c59 100644
--- a/common/common.h
+++ b/common/common.h
@@ -277,7 +277,8 @@ struct common_params {
     int32_t port           = 8080;         // server listens on this network port
     int32_t timeout_read   = 600;          // http read timeout in seconds
     int32_t timeout_write  = timeout_read; // http write timeout in seconds
-    int     n_threads_http = -1;           // number of threads to process HTTP requests (TODO: support threadpool)
+    int32_t n_threads_http = -1;           // number of threads to process HTTP requests (TODO: support threadpool)
+    int32_t n_cache_reuse  = 0;            // min chunk size to reuse from the cache via KV shifting
 
     std::string hostname      = "127.0.0.1";
     std::string public_path   = "";                                                                         // NOLINT
diff --git a/examples/server/README.md b/examples/server/README.md
index b5feeb77b..cd0eaf847 100644
--- a/examples/server/README.md
+++ b/examples/server/README.md
@@ -147,6 +147,7 @@ The project is under active development, and we are [looking for feedback and co
 | `--ssl-cert-file FNAME` | path to file a PEM-encoded SSL certificate
(env: LLAMA_ARG_SSL_CERT_FILE) |
 | `-to, --timeout N` | server read/write timeout in seconds (default: 600)
(env: LLAMA_ARG_TIMEOUT) |
 | `--threads-http N` | number of threads used to process HTTP requests (default: -1)
(env: LLAMA_ARG_THREADS_HTTP) |
+| `--cache-reuse N` | min chunk size to attempt reusing from the cache via KV shifting (default: 0)
(env: LLAMA_ARG_CACHE_REUSE) |
 | `--metrics` | enable prometheus compatible metrics endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_METRICS) |
 | `--slots` | enable slots monitoring endpoint (default: disabled)
(env: LLAMA_ARG_ENDPOINT_SLOTS) |
 | `--props` | enable changing global properties via POST /props (default: disabled)
(env: LLAMA_ARG_ENDPOINT_PROPS) |
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index f809c46d5..015b3b2c5 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -800,7 +800,7 @@ struct server_context {
                 int slot_prompt_len = slot_prompt.size();
 
                 // length of the Longest Common Prefix between the current slot's prompt and the input prompt
-                int lcp_len = common_part(slot_prompt, prompt);
+                int lcp_len = longest_common_prefix(slot_prompt, prompt);
 
                 // fraction of the common substring length compared to the current slot's prompt length
                 similarity = static_cast(lcp_len) / slot_prompt_len;
@@ -2012,7 +2012,7 @@ struct server_context {
                             }
                             slot.params.n_keep = std::min(slot.n_ctx - 4, slot.params.n_keep);
 
-                            // if input prompt is too big, truncate it (if group attention self-extend is disabled)
+                            // if input prompt is too big, truncate it
                             if (slot.n_prompt_tokens >= slot.n_ctx) {
                                 const int n_left = slot.n_ctx - slot.params.n_keep;
 
@@ -2042,12 +2042,74 @@ struct server_context {
 
                             if (slot.params.cache_prompt) {
                                 // reuse any previously computed tokens that are common with the new prompt
-                                slot.n_past = common_part(slot.cache_tokens, prompt_tokens);
+                                slot.n_past = longest_common_prefix(slot.cache_tokens, prompt_tokens);
 
                                 // push the prompt into the sampling context (do not apply grammar)
                                 for (int i = 0; i < slot.n_past; ++i) {
                                     common_sampler_accept(slot.smpl, slot.cache_tokens[i], false);
                                 }
+
+                                // reuse chunks from the cached prompt by shifting their KV cache in the new position
+                                if (params.n_cache_reuse > 0) {
+                                    size_t head_c = slot.n_past; // cache
+                                    size_t head_p = slot.n_past; // current prompt
+
+                                    SLT_DBG(slot, "trying to reuse chunks with size > %d, slot.n_past = %d\n", params.n_cache_reuse, slot.n_past);
+
+                                    while (head_c < slot.cache_tokens.size() &&
+                                           head_p < prompt_tokens.size()) {
+                                        if (llama_token_is_control(model, slot.cache_tokens[head_c])) {
+                                            break;
+                                        }
+
+                                        if (llama_token_is_control(model, prompt_tokens[head_p])) {
+                                            break;
+                                        }
+
+                                        size_t n_match = 0;
+
+                                        while (head_c + n_match < slot.cache_tokens.size() &&
+                                               head_p + n_match < prompt_tokens.size()     &&
+                                               slot.cache_tokens[head_c + n_match] == prompt_tokens[head_p + n_match]) {
+                                            if (llama_token_is_control(model, slot.cache_tokens[head_c + n_match])) {
+                                                break;
+                                            }
+
+                                            if (llama_token_is_control(model, prompt_tokens[head_p + n_match])) {
+                                                break;
+                                            }
+
+                                            n_match++;
+                                        }
+
+                                        if (n_match >= (size_t) params.n_cache_reuse) {
+                                            SLT_DBG(slot, "reusing chunk with size %zu, shifting KV cache [%zu, %zu) -> [%zu, %zu)\n", n_match, head_c, head_c + n_match, head_p, head_p + n_match);
+                                            //for (size_t i = head_p; i < head_p + n_match; i++) {
+                                            //    SLT_DBG(slot, "cache token %3zu: %6d '%s'\n", i, prompt_tokens[i], common_token_to_piece(ctx, prompt_tokens[i]).c_str());
+                                            //}
+
+                                            const int64_t kv_shift = (int64_t) head_p - (int64_t) head_c;
+
+                                            llama_kv_cache_seq_rm (ctx, slot.id + 1, head_p, head_c);
+                                            llama_kv_cache_seq_add(ctx, slot.id + 1, head_c, -1,     kv_shift);
+
+                                            for (size_t i = 0; i < n_match; i++) {
+                                                slot.cache_tokens[head_p + i] = slot.cache_tokens[head_c + i];
+
+                                                common_sampler_accept(slot.smpl, slot.cache_tokens[head_p + i], false);
+
+                                                slot.n_past++;
+                                            }
+
+                                            head_c += n_match;
+                                            head_p += n_match;
+                                        } else {
+                                            head_c += 1;
+                                        }
+                                    }
+
+                                    SLT_DBG(slot, "after context reuse, new slot.n_past = %d\n", slot.n_past);
+                                }
                             }
                         }
 
@@ -3257,6 +3319,7 @@ int main(int argc, char ** argv) {
 
     ctx_server.queue_tasks.on_new_task(std::bind(
                 &server_context::process_single_task, &ctx_server, std::placeholders::_1));
+
     ctx_server.queue_tasks.on_update_slots(std::bind(
                 &server_context::update_slots, &ctx_server));
 
diff --git a/examples/server/utils.hpp b/examples/server/utils.hpp
index ad99e9574..37999604d 100644
--- a/examples/server/utils.hpp
+++ b/examples/server/utils.hpp
@@ -195,14 +195,14 @@ static std::string gen_chatcmplid() {
 // other common utils
 //
 
-static size_t common_part(const std::vector & a, const std::vector & b) {
+static size_t longest_common_prefix(const std::vector & a, const std::vector & b) {
     size_t i;
     for (i = 0; i < a.size() && i < b.size() && a[i] == b[i]; i++) {}
 
     return i;
 }
 
-static size_t common_part(const std::string & a, const std::string & b) {
+static size_t longest_common_prefix(const std::string & a, const std::string & b) {
     size_t i;
     for (i = 0; i < a.size() && i < b.size() && a[i] == b[i]; i++) {}
 

From d4c19c0f5cdb1e512573e8c86c79e8d0238c73c4 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Sun, 13 Oct 2024 21:31:35 +0300
Subject: [PATCH 09/36] server : accept extra_context for the infill endpoint
 (#9874)

* server : accept extra_context for the infill endpoint

ggml-ci

* server : update readme [no ci]

* server : use repo-level FIM pattern if possible

ggml-ci
---
 examples/server/README.md  |  21 ++++++++
 examples/server/server.cpp | 102 ++++++++++++++++++++++++++++++++++---
 src/llama.cpp              |  56 +++++++++++++-------
 3 files changed, 153 insertions(+), 26 deletions(-)

diff --git a/examples/server/README.md b/examples/server/README.md
index cd0eaf847..eb0a7b32e 100644
--- a/examples/server/README.md
+++ b/examples/server/README.md
@@ -524,9 +524,30 @@ Takes a prefix and a suffix and returns the predicted completion as stream.
 
 - `input_prefix`: Set the prefix of the code to infill.
 - `input_suffix`: Set the suffix of the code to infill.
+- `prompt`: Added after the `FIM_MID` token
+- `extra_context`: Additional context inserted before the FIM prefix. See https://github.com/ggerganov/llama.cpp/pull/9874
 
 It also accepts all the options of `/completion`.
 
+If the model has `FIM_REPO` and `FIM_FILE_SEP` tokens, the [repo-level pattern](https://arxiv.org/pdf/2409.12186) is used:
+
+```txt
+myproject
+{chunk 0 filename}
+{chunk 0 text}
+{chunk 1 filename}
+{chunk 1 text}
+...
+filename
+[input_prefix][input_suffix][prompt]
+```
+
+If the tokens are missing, then the extra context is simply prefixed at the start:
+
+```txt
+[extra_context][input_prefix][input_suffix][prompt]
+```
+
 ### **GET** `/props`: Get server global properties.
 
 This endpoint is public (no API key check). By default, it is read-only. To make POST request to change global properties, you need to start server with `--props`
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index 015b3b2c5..18bcad3f0 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -139,6 +139,7 @@ struct slot_params {
 
     json input_prefix;
     json input_suffix;
+    json extra_context;
 };
 
 struct server_slot {
@@ -170,6 +171,7 @@ struct server_slot {
 
     // when a task is submitted, we first tokenize the prompt and store it here
     std::vector prompt_tokens;
+    std::vector extra_tokens;
 
     std::string generated_text;
     std::vector cache_tokens;
@@ -906,8 +908,26 @@ struct server_context {
         }
 
         // infill
-        slot.params.input_prefix = json_value(data, "input_prefix", default_params.input_prefix);
-        slot.params.input_suffix = json_value(data, "input_suffix", default_params.input_suffix);
+        slot.params.input_prefix  = json_value(data, "input_prefix",  default_params.input_prefix);
+        slot.params.input_suffix  = json_value(data, "input_suffix",  default_params.input_suffix);
+        slot.params.extra_context = json_value(data, "extra_context", default_params.extra_context);
+
+        SLT_DBG(slot, "extra_context chunks: %d\n", (int) slot.params.extra_context.size());
+        for (const auto & chunk : slot.params.extra_context) {
+            // { "text": string, "filename": string }
+            if (!chunk.contains("text") || !chunk["text"].is_string()) {
+                send_error(task, "extra_context chunk must contain a \"text\" field with a string value", ERROR_TYPE_INVALID_REQUEST);
+                return false;
+            }
+
+            // filename is optional
+            if (chunk.contains("filename") && !chunk["filename"].is_string()) {
+                send_error(task, "extra_context chunk's \"filename\" field must be a string", ERROR_TYPE_INVALID_REQUEST);
+                return false;
+            }
+
+            SLT_DBG(slot, "extra_context chunk in file '%s':\n%s\n", chunk.value("filename", "").c_str(), chunk.value("text", "").c_str());
+        }
 
         // get prompt
         if (task.cmpl_type != SERVER_TASK_CMPL_TYPE_INFILL) {
@@ -1934,13 +1954,66 @@ struct server_context {
                                 } break;
                             case SERVER_TASK_CMPL_TYPE_INFILL:
                                 {
+                                    // use FIM repo-level pattern:
+                                    // ref: https://arxiv.org/pdf/2409.12186
+                                    //
+                                    // [FIM_REP]myproject
+                                    // [FIM_SEP]filename0
+                                    // extra chunk 0
+                                    // [FIM_SEP]filename1
+                                    // extra chunk 1
+                                    // ...
+                                    // [FIM_SEP]filename
+                                    // [FIM_PRE]prefix[FIM_SUF]suffix[FIM_MID]
+                                    //
                                     auto prefix_tokens = tokenize(slot.params.input_prefix, false, false);
                                     auto suffix_tokens = tokenize(slot.params.input_suffix, false, false);
 
-                                    // for now pick context to fit in a single batch (ratio prefix:suffix = 3:1, TODO: configurable?)
-                                    const int n_suffix_take = std::min(suffix_tokens.size(), n_batch/4);
+                                    slot.extra_tokens.clear();
+                                    if (llama_token_fim_rep(model) != LLAMA_TOKEN_NULL) {
+                                        static const auto k_fim_repo = tokenize("myproject\n", false, false);
+
+                                        slot.extra_tokens.push_back(llama_token_fim_rep(model));
+                                        slot.extra_tokens.insert(slot.extra_tokens.end(), k_fim_repo.begin(), k_fim_repo.end());
+                                    }
+
+                                    for (const auto & chunk : slot.params.extra_context) {
+                                        // { "text": string, "filename": string }
+                                        const std::string text     = chunk.value("text", "");
+                                        const std::string filename = chunk.value("filename", "tmp");
+
+                                        if (llama_token_fim_sep(model) != LLAMA_TOKEN_NULL) {
+                                            const auto k_fim_file = tokenize(filename + "\n", false, false);
+
+                                            slot.extra_tokens.insert(slot.extra_tokens.end(), llama_token_fim_sep(model));
+                                            slot.extra_tokens.insert(slot.extra_tokens.end(), k_fim_file.begin(), k_fim_file.end());
+                                        } else {
+                                            // chunk separator in binary form to avoid confusing the AI
+                                            static const char k_chunk_prefix_str[] = {0x0a, 0x0a, 0x2d, 0x2d, 0x2d, 0x20, 0x73, 0x6e, 0x69, 0x70, 0x70, 0x65, 0x74, 0x20, 0x2d, 0x2d, 0x2d, 0x0a, 0x0a, 0x00};
+                                            static const auto k_chunk_prefix_tokens = tokenize(k_chunk_prefix_str, false, false);
+
+                                            slot.extra_tokens.insert(slot.extra_tokens.end(), k_chunk_prefix_tokens.begin(), k_chunk_prefix_tokens.end());
+                                        }
+
+                                        const auto chunk_tokens = tokenize(text, false, false);
+                                        slot.extra_tokens.insert(slot.extra_tokens.end(), chunk_tokens.begin(), chunk_tokens.end());
+                                    }
+
+                                    if (llama_token_fim_sep(model) != LLAMA_TOKEN_NULL) {
+                                        // TODO: current filename
+                                        static const auto k_fim_file = tokenize("filename\n", false, false);
+
+                                        slot.extra_tokens.insert(slot.extra_tokens.end(), llama_token_fim_sep(model));
+                                        slot.extra_tokens.insert(slot.extra_tokens.end(), k_fim_file.begin(), k_fim_file.end());
+                                    }
+
+                                    // for now pick FIM context to fit in a batch (ratio prefix:suffix = 3:1, TODO: configurable?)
+                                    const int n_suffix_take = std::min(suffix_tokens.size(), (n_batch)/4);
                                     const int n_prefix_take = std::min(prefix_tokens.size(), (n_batch - 3) - n_suffix_take);
 
+                                    // fill the rest of the context with extra chunks
+                                    const int n_extra_take = std::min(std::max(0, slot.n_ctx - (n_batch) - 2*slot.n_predict), slot.extra_tokens.size());
+
                                     prefix_tokens.erase(prefix_tokens.begin(), prefix_tokens.begin() + prefix_tokens.size() - n_prefix_take);
                                     suffix_tokens.resize(n_suffix_take);
 
@@ -1954,6 +2027,11 @@ struct server_context {
                                         embd_inp.insert(embd_inp.begin(), llama_token_bos(model));
                                     }
 
+                                    SLT_DBG(slot, "extra: n_ctx = %d, n_extra_take = %d, n_extra = %d\n", slot.n_ctx, n_extra_take, (int) slot.extra_tokens.size());
+
+                                    // put the extra context before the FIM prefix
+                                    embd_inp.insert(embd_inp.begin(), slot.extra_tokens.end() - n_extra_take, slot.extra_tokens.end());
+
                                     embd_inp.insert(embd_inp.end(), embd_end.begin(), embd_end.end());
                                     embd_inp.push_back(llama_token_fim_mid(model));
 
@@ -2058,11 +2136,15 @@ struct server_context {
 
                                     while (head_c < slot.cache_tokens.size() &&
                                            head_p < prompt_tokens.size()) {
-                                        if (llama_token_is_control(model, slot.cache_tokens[head_c])) {
+                                        if (llama_token_is_control(model, slot.cache_tokens[head_c]) &&
+                                            slot.cache_tokens[head_c] != llama_token_fim_rep(model) &&
+                                            slot.cache_tokens[head_c] != llama_token_fim_sep(model)) {
                                             break;
                                         }
 
-                                        if (llama_token_is_control(model, prompt_tokens[head_p])) {
+                                        if (llama_token_is_control(model, prompt_tokens[head_p]) &&
+                                            prompt_tokens[head_p] != llama_token_fim_rep(model) &&
+                                            prompt_tokens[head_p] != llama_token_fim_sep(model)) {
                                             break;
                                         }
 
@@ -2071,11 +2153,15 @@ struct server_context {
                                         while (head_c + n_match < slot.cache_tokens.size() &&
                                                head_p + n_match < prompt_tokens.size()     &&
                                                slot.cache_tokens[head_c + n_match] == prompt_tokens[head_p + n_match]) {
-                                            if (llama_token_is_control(model, slot.cache_tokens[head_c + n_match])) {
+                                            if (llama_token_is_control(model, slot.cache_tokens[head_c + n_match]) &&
+                                                slot.cache_tokens[head_c + n_match] != llama_token_fim_rep(model) &&
+                                                slot.cache_tokens[head_c + n_match] != llama_token_fim_sep(model)) {
                                                 break;
                                             }
 
-                                            if (llama_token_is_control(model, prompt_tokens[head_p + n_match])) {
+                                            if (llama_token_is_control(model, prompt_tokens[head_p + n_match]) &&
+                                                prompt_tokens[head_p + n_match] != llama_token_fim_rep(model) &&
+                                                prompt_tokens[head_p + n_match] != llama_token_fim_sep(model)) {
                                                 break;
                                             }
 
diff --git a/src/llama.cpp b/src/llama.cpp
index f68024f5b..511f91802 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -6596,8 +6596,8 @@ static void llm_load_vocab(
                    ) {
                     vocab.special_eot_id = t.second;
                     if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.first.c_str());
+                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.second, t.first.c_str());
                         vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                     }
                 }
@@ -6610,8 +6610,8 @@ static void llm_load_vocab(
                         ) {
                     vocab.special_eom_id = t.second;
                     if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.first.c_str());
+                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.second, t.first.c_str());
                         vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                     }
                 }
@@ -6627,8 +6627,8 @@ static void llm_load_vocab(
                         ) {
                     vocab.special_fim_pre_id = t.second;
                     if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.first.c_str());
+                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.second, t.first.c_str());
                         vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                     }
                 }
@@ -6644,8 +6644,8 @@ static void llm_load_vocab(
                         ) {
                     vocab.special_fim_suf_id = t.second;
                     if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.first.c_str());
+                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.second, t.first.c_str());
                         vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                     }
                 }
@@ -6661,8 +6661,8 @@ static void llm_load_vocab(
                         ) {
                     vocab.special_fim_mid_id = t.second;
                     if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.first.c_str());
+                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.second, t.first.c_str());
                         vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                     }
                 }
@@ -6677,8 +6677,8 @@ static void llm_load_vocab(
                         ) {
                     vocab.special_fim_pad_id = t.second;
                     if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.first.c_str());
+                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.second, t.first.c_str());
                         vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                     }
                 }
@@ -6694,8 +6694,8 @@ static void llm_load_vocab(
                         ) {
                     vocab.special_fim_rep_id = t.second;
                     if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.first.c_str());
+                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.second, t.first.c_str());
                         vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                     }
                 }
@@ -6708,8 +6708,8 @@ static void llm_load_vocab(
                         ) {
                     vocab.special_fim_sep_id = t.second;
                     if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                        LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                                __func__, t.first.c_str());
+                        LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                                __func__, t.second, t.first.c_str());
                         vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                     }
                 }
@@ -6720,6 +6720,19 @@ static void llm_load_vocab(
         // this is currently determined based on the token text, which is obviously not ideal
         // ref: https://github.com/ggerganov/llama.cpp/issues/9606
         vocab.special_eog_ids.clear();
+
+        if (vocab.special_fim_pad_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_fim_pad_id) == 0) {
+            vocab.special_eog_ids.insert(vocab.special_fim_pad_id);
+        }
+
+        if (vocab.special_fim_rep_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_fim_rep_id) == 0) {
+            vocab.special_eog_ids.insert(vocab.special_fim_rep_id);
+        }
+
+        if (vocab.special_fim_sep_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_fim_sep_id) == 0) {
+            vocab.special_eog_ids.insert(vocab.special_fim_sep_id);
+        }
+
         for (const auto & t : vocab.token_to_id) {
             if (false
                     || t.first == "<|eot_id|>"
@@ -6732,13 +6745,20 @@ static void llm_load_vocab(
                ) {
                 vocab.special_eog_ids.insert(t.second);
                 if ((vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
-                    LLAMA_LOG_WARN("%s: control-looking token: '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
-                            __func__, t.first.c_str());
+                    LLAMA_LOG_WARN("%s: control-looking token: %6d '%s' was not control-type; this is probably a bug in the model. its type will be overridden\n",
+                            __func__, t.second, t.first.c_str());
                     vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                 }
+            } else {
+                // token is control, but not marked as EOG -> print a warning
+                if (vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL && vocab.special_eog_ids.count(t.second) == 0) {
+                    LLAMA_LOG_WARN("%s: control token: %6d '%s' is not marked as EOG\n",
+                            __func__, t.second, t.first.c_str());
+                }
             }
         }
 
+        // sanity checks
         if (vocab.special_eos_id != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(vocab.special_eos_id) == 0) {
             vocab.special_eog_ids.insert(vocab.special_eos_id);
             LLAMA_LOG_WARN("%s: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);

From 13dca2a54a394757d56fdd652b9f0df08f44ea22 Mon Sep 17 00:00:00 2001
From: agray3 
Date: Mon, 14 Oct 2024 01:49:08 +0100
Subject: [PATCH 10/36] Vectorize load instructions in dmmv f16 CUDA kernel
 (#9816)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

* Vectorize load instructions in dmmv f16 CUDA kernel

Replaces scalar with vector load instructions, which substantially
improves performance on NVIDIA HBM GPUs, e.g. gives a 1.27X overall
speedup for Meta-Llama-3-8B-Instruct-F16 BS1 inference evaluation on
H100 SXM 80GB HBM3. On GDDR GPUs, there is a slight (1.01X) speedup.

* addressed comment

* Update ggml/src/ggml-cuda/dmmv.cu

Co-authored-by: Johannes Gäßler 

---------

Co-authored-by: Johannes Gäßler 
---
 ggml/src/ggml-cuda/dmmv.cu | 34 +++++++++++++++++++++++++---------
 1 file changed, 25 insertions(+), 9 deletions(-)

diff --git a/ggml/src/ggml-cuda/dmmv.cu b/ggml/src/ggml-cuda/dmmv.cu
index 96a5adef5..00e21b5d7 100644
--- a/ggml/src/ggml-cuda/dmmv.cu
+++ b/ggml/src/ggml-cuda/dmmv.cu
@@ -416,10 +416,11 @@ static __global__ void dequantize_mul_mat_vec_q6_k(const void * __restrict__ vx,
 
 static __device__ void convert_f16(const void * vx, const int64_t ib, const int iqs, dfloat2 & v){
     const half * x = (const half *) vx;
-
+    // load 2 halfs into register in a single instruction
+    const half2 x_reg = *((half2 *) &(x[ib + iqs]));
     // automatic half -> float type cast if dfloat == float
-    v.x = x[ib + iqs + 0];
-    v.y = x[ib + iqs + 1];
+    v.x = __low2float(x_reg);
+    v.y = __high2float(x_reg);
 }
 
 static constexpr __device__ dequantize_kernel_t get_dequantize_kernel(ggml_type type) {
@@ -476,13 +477,28 @@ static __global__ void dequantize_mul_mat_vec(const void * __restrict__ vx, cons
             // matrix multiplication
             // for qr = 2 the y index needs to increase by 1 per j iter because of y_offset = qk/2
 #ifdef GGML_CUDA_F16
-            tmp += __hmul2(v, {
-                y[iybs + iqs + j/qr + 0],
-                y[iybs + iqs + j/qr + y_offset]
-            });
+            if ( y_offset == 1 ) {
+                // load 2 dfloats into register in a single instruction
+                const dfloat2 y_reg = *((dfloat2 *) &(y[iybs + iqs + j/qr]));
+                tmp += __hmul2(v, y_reg);
+            }
+            else {
+                tmp += __hmul2(v, {
+                        y[iybs + iqs + j/qr + 0],
+                        y[iybs + iqs + j/qr + y_offset]
+                    });
+            }
 #else
-            tmp += v.x * y[iybs + iqs + j/qr + 0];
-            tmp += v.y * y[iybs + iqs + j/qr + y_offset];
+            if ( y_offset == 1 ) {
+                // load 2 dfloats into register in a single instruction
+                const dfloat2 y_reg = *((dfloat2 *) &(y[iybs + iqs + j/qr]));
+                tmp += v.x * y_reg.x;
+                tmp += v.y * y_reg.y;
+            }
+            else {
+                tmp += v.x * y[iybs + iqs + j/qr + 0];
+                tmp += v.y * y[iybs + iqs + j/qr + y_offset];
+            }
 #endif // GGML_CUDA_F16
         }
     }

From a89f75e1b7b90cb2d4d4c52ca53ef9e9b466aa45 Mon Sep 17 00:00:00 2001
From: VoidIsVoid <343750470@qq.com>
Date: Mon, 14 Oct 2024 15:04:36 +0800
Subject: [PATCH 11/36] server : handle "logprobs" field with false value
 (#9871)

Co-authored-by: Gimling 
---
 examples/server/utils.hpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/examples/server/utils.hpp b/examples/server/utils.hpp
index 37999604d..69519ef95 100644
--- a/examples/server/utils.hpp
+++ b/examples/server/utils.hpp
@@ -360,9 +360,9 @@ static json oaicompat_completion_params_parse(
 
     // Handle "logprobs" field
     // TODO: The response format of this option is not yet OAI-compatible, but seems like no one really using it; We may need to fix it in the future
-    if (body.contains("logprobs")) {
+    if (json_value(body, "logprobs", false)) {
         llama_params["n_probs"] = json_value(body, "top_logprobs", 20);
-    } else if (body.contains("top_logprobs")) {
+    } else if (body.contains("top_logprobs") && !body.at("top_logprobs").is_null()) {
         throw std::runtime_error("top_logprobs requires logprobs to be set to true");
     }
 

From 4c42f93b22146c83b763d8cbee5fafc512746649 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Micha=C5=82=20Tuszy=C5=84ski?= 
Date: Tue, 15 Oct 2024 10:20:34 +0200
Subject: [PATCH 12/36] readme : update bindings list (#9889)

---
 README.md | 1 +
 1 file changed, 1 insertion(+)

diff --git a/README.md b/README.md
index dd4927b04..08fe8cc92 100644
--- a/README.md
+++ b/README.md
@@ -130,6 +130,7 @@ Typically finetunes of the base models below are supported as well.
 - Flutter/Dart: [netdur/llama_cpp_dart](https://github.com/netdur/llama_cpp_dart)
 - PHP (API bindings and features built on top of llama.cpp): [distantmagic/resonance](https://github.com/distantmagic/resonance) [(more info)](https://github.com/ggerganov/llama.cpp/pull/6326)
 - Guile Scheme: [guile_llama_cpp](https://savannah.nongnu.org/projects/guile-llama-cpp)
+- Swift [srgtuszy/llama-cpp-swift](https://github.com/srgtuszy/llama-cpp-swift)
 
 **UI:**
 

From dcdd535302fc9702a4709be25f56540d65163a44 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Tue, 15 Oct 2024 12:48:44 +0300
Subject: [PATCH 13/36] server : update preact (#9895)

---
 examples/server/public/index.js | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/examples/server/public/index.js b/examples/server/public/index.js
index fe615ca25..32ec6e9e1 100644
--- a/examples/server/public/index.js
+++ b/examples/server/public/index.js
@@ -1 +1 @@
-const t=Symbol.for("preact-signals");function n(){if(r>1){r--;return}let t,n=!1;while(void 0!==i){let _=i;i=void 0;u++;while(void 0!==_){const i=_.o;_.o=void 0;_.f&=-3;if(!(8&_.f)&&h(_))try{_.c()}catch(e){if(!n){t=e;n=!0}}_=i}}u=0;r--;if(n)throw t}function e(t){if(r>0)return t();r++;try{return t()}finally{n()}}let _,i;function o(t){const n=_;_=void 0;try{return t()}finally{_=n}}let r=0,u=0,l=0;function f(t){if(void 0===_)return;let n=t.n;if(void 0===n||n.t!==_){n={i:0,S:t,p:_.s,n:void 0,t:_,e:void 0,x:void 0,r:n};if(void 0!==_.s)_.s.n=n;_.s=n;t.n=n;if(32&_.f)t.S(n);return n}else if(-1===n.i){n.i=0;if(void 0!==n.n){n.n.p=n.p;if(void 0!==n.p)n.p.n=n.n;n.p=_.s;n.n=void 0;_.s.n=n;_.s=n}return n}}function s(t){this.v=t;this.i=0;this.n=void 0;this.t=void 0}s.prototype.brand=t;s.prototype.h=function(){return!0};s.prototype.S=function(t){if(this.t!==t&&void 0===t.e){t.x=this.t;if(void 0!==this.t)this.t.e=t;this.t=t}};s.prototype.U=function(t){if(void 0!==this.t){const n=t.e,e=t.x;if(void 0!==n){n.x=e;t.e=void 0}if(void 0!==e){e.e=n;t.x=void 0}if(t===this.t)this.t=e}};s.prototype.subscribe=function(t){return k(()=>{const n=this.value,e=_;_=void 0;try{t(n)}finally{_=e}})};s.prototype.valueOf=function(){return this.value};s.prototype.toString=function(){return this.value+""};s.prototype.toJSON=function(){return this.value};s.prototype.peek=function(){const t=_;_=void 0;try{return this.value}finally{_=t}};Object.defineProperty(s.prototype,"value",{get(){const t=f(this);if(void 0!==t)t.i=this.i;return this.v},set(t){if(t!==this.v){if(u>100)throw new Error("Cycle detected");this.v=t;this.i++;l++;r++;try{for(let t=this.t;void 0!==t;t=t.x)t.t.N()}finally{n()}}}});function c(t){return new s(t)}function h(t){for(let n=t.s;void 0!==n;n=n.n)if(n.S.i!==n.i||!n.S.h()||n.S.i!==n.i)return!0;return!1}function a(t){for(let n=t.s;void 0!==n;n=n.n){const e=n.S.n;if(void 0!==e)n.r=e;n.S.n=n;n.i=-1;if(void 0===n.n){t.s=n;break}}}function p(t){let n,e=t.s;while(void 0!==e){const t=e.p;if(-1===e.i){e.S.U(e);if(void 0!==t)t.n=e.n;if(void 0!==e.n)e.n.p=t}else n=e;e.S.n=e.r;if(void 0!==e.r)e.r=void 0;e=t}t.s=n}function d(t){s.call(this,void 0);this.x=t;this.s=void 0;this.g=l-1;this.f=4}(d.prototype=new s).h=function(){this.f&=-3;if(1&this.f)return!1;if(32==(36&this.f))return!0;this.f&=-5;if(this.g===l)return!0;this.g=l;this.f|=1;if(this.i>0&&!h(this)){this.f&=-2;return!0}const t=_;try{a(this);_=this;const t=this.x();if(16&this.f||this.v!==t||0===this.i){this.v=t;this.f&=-17;this.i++}}catch(t){this.v=t;this.f|=16;this.i++}_=t;p(this);this.f&=-2;return!0};d.prototype.S=function(t){if(void 0===this.t){this.f|=36;for(let t=this.s;void 0!==t;t=t.n)t.S.S(t)}s.prototype.S.call(this,t)};d.prototype.U=function(t){if(void 0!==this.t){s.prototype.U.call(this,t);if(void 0===this.t){this.f&=-33;for(let t=this.s;void 0!==t;t=t.n)t.S.U(t)}}};d.prototype.N=function(){if(!(2&this.f)){this.f|=6;for(let t=this.t;void 0!==t;t=t.x)t.t.N()}};Object.defineProperty(d.prototype,"value",{get(){if(1&this.f)throw new Error("Cycle detected");const t=f(this);this.h();if(void 0!==t)t.i=this.i;if(16&this.f)throw this.v;return this.v}});function v(t){return new d(t)}function y(t){const e=t.u;t.u=void 0;if("function"==typeof e){r++;const i=_;_=void 0;try{e()}catch(n){t.f&=-2;t.f|=8;m(t);throw n}finally{_=i;n()}}}function m(t){for(let n=t.s;void 0!==n;n=n.n)n.S.U(n);t.x=void 0;t.s=void 0;y(t)}function g(t){if(_!==this)throw new Error("Out-of-order effect");p(this);_=t;this.f&=-2;if(8&this.f)m(this);n()}function b(t){this.x=t;this.u=void 0;this.s=void 0;this.o=void 0;this.f=32}b.prototype.c=function(){const t=this.S();try{if(8&this.f)return;if(void 0===this.x)return;const n=this.x();if("function"==typeof n)this.u=n}finally{t()}};b.prototype.S=function(){if(1&this.f)throw new Error("Cycle detected");this.f|=1;this.f&=-9;y(this);a(this);r++;const t=_;_=this;return g.bind(this,t)};b.prototype.N=function(){if(!(2&this.f)){this.f|=2;this.o=i;i=this}};b.prototype.d=function(){this.f|=8;if(!(1&this.f))m(this)};function k(t){const n=new b(t);try{n.c()}catch(t){n.d();throw t}return n.d.bind(n)}var w,S,x,C,U,E,H,P,N,$,T,D,M={},F=[],A=/acit|ex(?:s|g|n|p|$)|rph|grid|ows|mnc|ntw|ine[ch]|zoo|^ord|itera/i,W=Array.isArray;function L(t,n){for(var e in n)t[e]=n[e];return t}function O(t){var n=t.parentNode;n&&n.removeChild(t)}function R(t,n,e){var _,i,o,r={};for(o in n)"key"==o?_=n[o]:"ref"==o?i=n[o]:r[o]=n[o];if(arguments.length>2&&(r.children=arguments.length>3?w.call(arguments,2):e),"function"==typeof t&&null!=t.defaultProps)for(o in t.defaultProps)void 0===r[o]&&(r[o]=t.defaultProps[o]);return I(t,r,_,i,null)}function I(t,n,e,_,i){var o={type:t,props:n,key:e,ref:_,__k:null,__:null,__b:0,__e:null,__d:void 0,__c:null,constructor:void 0,__v:null==i?++x:i,__i:-1,__u:0};return null==i&&null!=S.vnode&&S.vnode(o),o}function V(){return{current:null}}function j(t){return t.children}function q(t,n){this.props=t,this.context=n}function B(t,n){if(null==n)return t.__?B(t.__,t.__i+1):null;for(var e;nn&&U.sort(P));J.__r=0}function K(t,n,e,_,i,o,r,u,l,f,s){var c,h,a,p,d,v=_&&_.__k||F,y=n.length;for(e.__d=l,Q(e,n,v),l=e.__d,c=0;c0?I(i.type,i.props,i.key,i.ref?i.ref:null,i.__v):i)?(i.__=t,i.__b=t.__b+1,u=Z(i,e,r,s),i.__i=u,o=null,-1!==u&&(s--,(o=e[u])&&(o.__u|=131072)),null==o||null===o.__v?(-1==u&&c--,"function"!=typeof i.type&&(i.__u|=65536)):u!==r&&(u==r-1?c--:u==r+1?c++:u>r?s>l-r?c+=u-r:c--:u(null!=l&&0==(131072&l.__u)?1:0))for(;r>=0||u=0){if((l=n[r])&&0==(131072&l.__u)&&i==l.key&&o===l.type)return r;r--}if(u2&&(u.children=arguments.length>3?w.call(arguments,2):e),I(t.type,u,_||t.key,i||t.ref,null)}function ht(t,n){var e={__c:n="__cC"+D++,__:t,Consumer:function(t,n){return t.children(n)},Provider:function(t){var e,_;return this.getChildContext||(e=[],(_={})[n]=this,this.getChildContext=function(){return _},this.componentWillUnmount=function(){e=null},this.shouldComponentUpdate=function(t){this.props.value!==t.value&&e.some((function(t){t.__e=!0,G(t)}))},this.sub=function(t){e.push(t);var n=t.componentWillUnmount;t.componentWillUnmount=function(){e&&e.splice(e.indexOf(t),1),n&&n.call(t)}}),t.children}};return e.Provider.__=e.Consumer.contextType=e}w=F.slice,S={__e:function(t,n,e,_){for(var i,o,r;n=n.__;)if((i=n.__c)&&!i.__)try{if((o=i.constructor)&&null!=o.getDerivedStateFromError&&(i.setState(o.getDerivedStateFromError(t)),r=i.__d),null!=i.componentDidCatch&&(i.componentDidCatch(t,_||{}),r=i.__d),r)return i.__E=i}catch(n){t=n}throw t}},x=0,C=function(t){return null!=t&&null==t.constructor},q.prototype.setState=function(t,n){var e;e=null!=this.__s&&this.__s!==this.state?this.__s:this.__s=L({},this.state),"function"==typeof t&&(t=t(L({},e),this.props)),t&&L(e,t),null!=t&&this.__v&&(n&&this._sb.push(n),G(this))},q.prototype.forceUpdate=function(t){this.__v&&(this.__e=!0,t&&this.__h.push(t),G(this))},q.prototype.render=j,U=[],H="function"==typeof Promise?Promise.prototype.then.bind(Promise.resolve()):setTimeout,P=function(t,n){return t.__v.__b-n.__v.__b},J.__r=0,N=0,$=et(!1),T=et(!0),D=0;var at,pt,dt,vt,yt=0,mt=[],gt=S,bt=gt.__b,kt=gt.__r,wt=gt.diffed,St=gt.__c,xt=gt.unmount,Ct=gt.__;function Ut(t,n){gt.__h&>.__h(pt,t,yt||n),yt=0;var e=pt.__H||(pt.__H={__:[],__h:[]});return t>=e.__.length&&e.__.push({}),e.__[t]}function Et(t){return yt=1,Ht(Bt,t)}function Ht(t,n,e){var _=Ut(at++,2);if(_.t=t,!_.__c&&(_.__=[e?e(n):Bt(void 0,n),function(t){var n=_.__N?_.__N[0]:_.__[0],e=_.t(n,t);n!==e&&(_.__N=[e,_.__[1]],_.__c.setState({}))}],_.__c=pt,!pt.u)){var i=function(t,n,e){if(!_.__c.__H)return!0;var i=_.__c.__H.__.filter((function(t){return!!t.__c}));if(i.every((function(t){return!t.__N})))return!o||o.call(this,t,n,e);var r=!1;return i.forEach((function(t){if(t.__N){var n=t.__[0];t.__=t.__N,t.__N=void 0,n!==t.__[0]&&(r=!0)}})),!(!r&&_.__c.props===t)&&(!o||o.call(this,t,n,e))};pt.u=!0;var o=pt.shouldComponentUpdate,r=pt.componentWillUpdate;pt.componentWillUpdate=function(t,n,e){if(this.__e){var _=o;o=void 0,i(t,n,e),o=_}r&&r.call(this,t,n,e)},pt.shouldComponentUpdate=i}return _.__N||_.__}function Pt(t,n){var e=Ut(at++,3);!gt.__s&&qt(e.__H,n)&&(e.__=t,e.i=n,pt.__H.__h.push(e))}function Nt(t,n){var e=Ut(at++,4);!gt.__s&&qt(e.__H,n)&&(e.__=t,e.i=n,pt.__h.push(e))}function $t(t){return yt=5,Dt((function(){return{current:t}}),[])}function Tt(t,n,e){yt=6,Nt((function(){return"function"==typeof t?(t(n()),function(){return t(null)}):t?(t.current=n(),function(){return t.current=null}):void 0}),null==e?e:e.concat(t))}function Dt(t,n){var e=Ut(at++,7);return qt(e.__H,n)&&(e.__=t(),e.__H=n,e.__h=t),e.__}function Mt(t,n){return yt=8,Dt((function(){return t}),n)}function Ft(t){var n=pt.context[t.__c],e=Ut(at++,9);return e.c=t,n?(null==e.__&&(e.__=!0,n.sub(pt)),n.props.value):t.__}function At(t,n){gt.useDebugValue&>.useDebugValue(n?n(t):t)}function Wt(t){var n=Ut(at++,10),e=Et();return n.__=t,pt.componentDidCatch||(pt.componentDidCatch=function(t,_){n.__&&n.__(t,_),e[1](t)}),[e[0],function(){e[1](void 0)}]}function Lt(){var t=Ut(at++,11);if(!t.__){for(var n=pt.__v;null!==n&&!n.__m&&null!==n.__;)n=n.__;var e=n.__m||(n.__m=[0,0]);t.__="P"+e[0]+"-"+e[1]++}return t.__}function Ot(){for(var t;t=mt.shift();)if(t.__P&&t.__H)try{t.__H.__h.forEach(Vt),t.__H.__h.forEach(jt),t.__H.__h=[]}catch(n){t.__H.__h=[],gt.__e(n,t.__v)}}gt.__b=function(t){pt=null,bt&&bt(t)},gt.__=function(t,n){t&&n.__k&&n.__k.__m&&(t.__m=n.__k.__m),Ct&&Ct(t,n)},gt.__r=function(t){kt&&kt(t),at=0;var n=(pt=t.__c).__H;n&&(dt===pt?(n.__h=[],pt.__h=[],n.__.forEach((function(t){t.__N&&(t.__=t.__N),t.i=t.__N=void 0}))):(n.__h.forEach(Vt),n.__h.forEach(jt),n.__h=[],at=0)),dt=pt},gt.diffed=function(t){wt&&wt(t);var n=t.__c;n&&n.__H&&(n.__H.__h.length&&(1!==mt.push(n)&&vt===gt.requestAnimationFrame||((vt=gt.requestAnimationFrame)||It)(Ot)),n.__H.__.forEach((function(t){t.i&&(t.__H=t.i),t.i=void 0}))),dt=pt=null},gt.__c=function(t,n){n.some((function(t){try{t.__h.forEach(Vt),t.__h=t.__h.filter((function(t){return!t.__||jt(t)}))}catch(r){n.some((function(t){t.__h&&(t.__h=[])})),n=[],gt.__e(r,t.__v)}})),St&&St(t,n)},gt.unmount=function(t){xt&&xt(t);var n,e=t.__c;e&&e.__H&&(e.__H.__.forEach((function(t){try{Vt(t)}catch(t){n=t}})),e.__H=void 0,n&>.__e(n,e.__v))};var Rt="function"==typeof requestAnimationFrame;function It(t){var n,e=function(){clearTimeout(_),Rt&&cancelAnimationFrame(n),setTimeout(t)},_=setTimeout(e,100);Rt&&(n=requestAnimationFrame(e))}function Vt(t){var n=pt,e=t.__c;"function"==typeof e&&(t.__c=void 0,e()),pt=n}function jt(t){var n=pt;t.__c=t.__(),pt=n}function qt(t,n){return!t||t.length!==n.length||n.some((function(n,e){return n!==t[e]}))}function Bt(t,n){return"function"==typeof n?n(t):n}function zt(t,n){S[t]=n.bind(null,S[t]||(()=>{}))}let Gt,Jt;function Kt(t){if(Jt)Jt();Jt=t&&t.S()}function Qt({data:t}){const n=Yt(t);n.value=t;const e=Dt(()=>{let t=this.__v;while(t=t.__)if(t.__c){t.__c.__$f|=4;break}this.__$u.c=()=>{var t;if(!C(e.peek())&&3===(null==(t=this.base)?void 0:t.nodeType))this.base.data=e.peek();else{this.__$f|=1;this.setState({})}};return v(()=>{let t=n.value.value;return 0===t?0:!0===t?"":t||""})},[]);return e.value}Qt.displayName="_st";Object.defineProperties(s.prototype,{constructor:{configurable:!0,value:void 0},type:{configurable:!0,value:Qt},props:{configurable:!0,get(){return{data:this}}},__b:{configurable:!0,value:1}});zt("__b",(t,n)=>{if("string"==typeof n.type){let t,e=n.props;for(let _ in e){if("children"===_)continue;let i=e[_];if(i instanceof s){if(!t)n.__np=t={};t[_]=i;e[_]=i.peek()}}}t(n)});zt("__r",(t,n)=>{Kt();let e,_=n.__c;if(_){_.__$f&=-2;e=_.__$u;if(void 0===e)_.__$u=e=function(t){let n;k((function(){n=this}));n.c=()=>{_.__$f|=1;_.setState({})};return n}()}Gt=_;Kt(e);t(n)});zt("__e",(t,n,e,_)=>{Kt();Gt=void 0;t(n,e,_)});zt("diffed",(t,n)=>{Kt();Gt=void 0;let e;if("string"==typeof n.type&&(e=n.__e)){let t=n.__np,_=n.props;if(t){let n=e.U;if(n)for(let e in n){let _=n[e];if(void 0!==_&&!(e in t)){_.d();n[e]=void 0}}else{n={};e.U=n}for(let i in t){let o=n[i],r=t[i];if(void 0===o){o=Xt(e,i,r,_);n[i]=o}else o.o(r,_)}}}t(n)});function Xt(t,n,e,_){const i=n in t&&void 0===t.ownerSVGElement,o=c(e);return{o:(t,n)=>{o.value=t;_=n},d:k(()=>{const e=o.value.value;if(_[n]!==e){_[n]=e;if(i)t[n]=e;else if(e)t.setAttribute(n,e);else t.removeAttribute(n)}})}}zt("unmount",(t,n)=>{if("string"==typeof n.type){let t=n.__e;if(t){const n=t.U;if(n){t.U=void 0;for(let t in n){let e=n[t];if(e)e.d()}}}}else{let t=n.__c;if(t){const n=t.__$u;if(n){t.__$u=void 0;n.d()}}}t(n)});zt("__h",(t,n,e,_)=>{if(_<3||9===_)n.__$f|=2;t(n,e,_)});q.prototype.shouldComponentUpdate=function(t,n){const e=this.__$u;if(!(e&&void 0!==e.s||4&this.__$f))return!0;if(3&this.__$f)return!0;for(let _ in n)return!0;for(let _ in t)if("__source"!==_&&t[_]!==this.props[_])return!0;for(let _ in this.props)if(!(_ in t))return!0;return!1};function Yt(t){return Dt(()=>c(t),[])}function Zt(t){const n=$t(t);n.current=t;Gt.__$f|=4;return Dt(()=>v(()=>n.current()),[])}function tn(t){const n=$t(t);n.current=t;Pt(()=>k(()=>n.current()),[])}var nn=function(t,n,e,_){var i;n[0]=0;for(var o=1;o=5&&((i||!t&&5===_)&&(r.push(_,0,i,e),_=6),t&&(r.push(_,t,0,e),_=6)),i=""},l=0;l"===n?(_=1,i=""):i=n+i[0]:o?n===o?o="":i+=n:'"'===n||"'"===n?o=n:">"===n?(u(),_=1):_&&("="===n?(_=5,e=i,i=""):"/"===n&&(_<5||">"===t[l][f+1])?(u(),3===_&&(r=r[0]),_=r,(r=r[0]).push(2,0,_),_=0):" "===n||"\t"===n||"\n"===n||"\r"===n?(u(),_=2):i+=n),3===_&&"!--"===i&&(_=4,r=r[0])}return u(),r}(t)),n),arguments,[])).length>1?n:n[0]}var on=_n.bind(R);export{q as Component,j as Fragment,s as Signal,e as batch,ct as cloneElement,v as computed,ht as createContext,R as createElement,V as createRef,k as effect,R as h,on as html,st as hydrate,C as isValidElement,S as options,ft as render,c as signal,Y as toChildArray,o as untracked,Mt as useCallback,Zt as useComputed,Ft as useContext,At as useDebugValue,Pt as useEffect,Wt as useErrorBoundary,Lt as useId,Tt as useImperativeHandle,Nt as useLayoutEffect,Dt as useMemo,Ht as useReducer,$t as useRef,Yt as useSignal,tn as useSignalEffect,Et as useState};
+const t=Symbol.for("preact-signals");function n(){if(r>1){r--;return}let t,n=!1;while(void 0!==i){let _=i;i=void 0;u++;while(void 0!==_){const i=_.o;_.o=void 0;_.f&=-3;if(!(8&_.f)&&h(_))try{_.c()}catch(e){if(!n){t=e;n=!0}}_=i}}u=0;r--;if(n)throw t}function e(t){if(r>0)return t();r++;try{return t()}finally{n()}}let _,i;function o(t){const n=_;_=void 0;try{return t()}finally{_=n}}let r=0,u=0,l=0;function s(t){if(void 0===_)return;let n=t.n;if(void 0===n||n.t!==_){n={i:0,S:t,p:_.s,n:void 0,t:_,e:void 0,x:void 0,r:n};if(void 0!==_.s)_.s.n=n;_.s=n;t.n=n;if(32&_.f)t.S(n);return n}else if(-1===n.i){n.i=0;if(void 0!==n.n){n.n.p=n.p;if(void 0!==n.p)n.p.n=n.n;n.p=_.s;n.n=void 0;_.s.n=n;_.s=n}return n}}function f(t){this.v=t;this.i=0;this.n=void 0;this.t=void 0}f.prototype.brand=t;f.prototype.h=function(){return!0};f.prototype.S=function(t){if(this.t!==t&&void 0===t.e){t.x=this.t;if(void 0!==this.t)this.t.e=t;this.t=t}};f.prototype.U=function(t){if(void 0!==this.t){const n=t.e,e=t.x;if(void 0!==n){n.x=e;t.e=void 0}if(void 0!==e){e.e=n;t.x=void 0}if(t===this.t)this.t=e}};f.prototype.subscribe=function(t){return k(()=>{const n=this.value,e=_;_=void 0;try{t(n)}finally{_=e}})};f.prototype.valueOf=function(){return this.value};f.prototype.toString=function(){return this.value+""};f.prototype.toJSON=function(){return this.value};f.prototype.peek=function(){const t=_;_=void 0;try{return this.value}finally{_=t}};Object.defineProperty(f.prototype,"value",{get(){const t=s(this);if(void 0!==t)t.i=this.i;return this.v},set(t){if(t!==this.v){if(u>100)throw new Error("Cycle detected");this.v=t;this.i++;l++;r++;try{for(let t=this.t;void 0!==t;t=t.x)t.t.N()}finally{n()}}}});function c(t){return new f(t)}function h(t){for(let n=t.s;void 0!==n;n=n.n)if(n.S.i!==n.i||!n.S.h()||n.S.i!==n.i)return!0;return!1}function a(t){for(let n=t.s;void 0!==n;n=n.n){const e=n.S.n;if(void 0!==e)n.r=e;n.S.n=n;n.i=-1;if(void 0===n.n){t.s=n;break}}}function p(t){let n,e=t.s;while(void 0!==e){const t=e.p;if(-1===e.i){e.S.U(e);if(void 0!==t)t.n=e.n;if(void 0!==e.n)e.n.p=t}else n=e;e.S.n=e.r;if(void 0!==e.r)e.r=void 0;e=t}t.s=n}function d(t){f.call(this,void 0);this.x=t;this.s=void 0;this.g=l-1;this.f=4}(d.prototype=new f).h=function(){this.f&=-3;if(1&this.f)return!1;if(32==(36&this.f))return!0;this.f&=-5;if(this.g===l)return!0;this.g=l;this.f|=1;if(this.i>0&&!h(this)){this.f&=-2;return!0}const t=_;try{a(this);_=this;const t=this.x();if(16&this.f||this.v!==t||0===this.i){this.v=t;this.f&=-17;this.i++}}catch(t){this.v=t;this.f|=16;this.i++}_=t;p(this);this.f&=-2;return!0};d.prototype.S=function(t){if(void 0===this.t){this.f|=36;for(let t=this.s;void 0!==t;t=t.n)t.S.S(t)}f.prototype.S.call(this,t)};d.prototype.U=function(t){if(void 0!==this.t){f.prototype.U.call(this,t);if(void 0===this.t){this.f&=-33;for(let t=this.s;void 0!==t;t=t.n)t.S.U(t)}}};d.prototype.N=function(){if(!(2&this.f)){this.f|=6;for(let t=this.t;void 0!==t;t=t.x)t.t.N()}};Object.defineProperty(d.prototype,"value",{get(){if(1&this.f)throw new Error("Cycle detected");const t=s(this);this.h();if(void 0!==t)t.i=this.i;if(16&this.f)throw this.v;return this.v}});function v(t){return new d(t)}function y(t){const e=t.u;t.u=void 0;if("function"==typeof e){r++;const i=_;_=void 0;try{e()}catch(n){t.f&=-2;t.f|=8;m(t);throw n}finally{_=i;n()}}}function m(t){for(let n=t.s;void 0!==n;n=n.n)n.S.U(n);t.x=void 0;t.s=void 0;y(t)}function g(t){if(_!==this)throw new Error("Out-of-order effect");p(this);_=t;this.f&=-2;if(8&this.f)m(this);n()}function b(t){this.x=t;this.u=void 0;this.s=void 0;this.o=void 0;this.f=32}b.prototype.c=function(){const t=this.S();try{if(8&this.f)return;if(void 0===this.x)return;const n=this.x();if("function"==typeof n)this.u=n}finally{t()}};b.prototype.S=function(){if(1&this.f)throw new Error("Cycle detected");this.f|=1;this.f&=-9;y(this);a(this);r++;const t=_;_=this;return g.bind(this,t)};b.prototype.N=function(){if(!(2&this.f)){this.f|=2;this.o=i;i=this}};b.prototype.d=function(){this.f|=8;if(!(1&this.f))m(this)};function k(t){const n=new b(t);try{n.c()}catch(t){n.d();throw t}return n.d.bind(n)}var w,S,x,C,U,E,H,P,N,$,T,D,M={},A=[],F=/acit|ex(?:s|g|n|p|$)|rph|grid|ows|mnc|ntw|ine[ch]|zoo|^ord|itera/i,W=Array.isArray;function L(t,n){for(var e in n)t[e]=n[e];return t}function O(t){t&&t.parentNode&&t.parentNode.removeChild(t)}function R(t,n,e){var _,i,o,r={};for(o in n)"key"==o?_=n[o]:"ref"==o?i=n[o]:r[o]=n[o];if(arguments.length>2&&(r.children=arguments.length>3?w.call(arguments,2):e),"function"==typeof t&&null!=t.defaultProps)for(o in t.defaultProps)void 0===r[o]&&(r[o]=t.defaultProps[o]);return I(t,r,_,i,null)}function I(t,n,e,_,i){var o={type:t,props:n,key:e,ref:_,__k:null,__:null,__b:0,__e:null,__d:void 0,__c:null,constructor:void 0,__v:null==i?++x:i,__i:-1,__u:0};return null==i&&null!=S.vnode&&S.vnode(o),o}function V(){return{current:null}}function j(t){return t.children}function q(t,n){this.props=t,this.context=n}function B(t,n){if(null==n)return t.__?B(t.__,t.__i+1):null;for(var e;nn&&U.sort(P));J.__r=0}function K(t,n,e,_,i,o,r,u,l,s,f){var c,h,a,p,d,v=_&&_.__k||A,y=n.length;for(e.__d=l,Q(e,n,v),l=e.__d,c=0;c0?I(i.type,i.props,i.key,i.ref?i.ref:null,i.__v):i).__=t,i.__b=t.__b+1,o=null,-1!==(u=i.__i=Z(i,e,r,f))&&(f--,(o=e[u])&&(o.__u|=131072)),null==o||null===o.__v?(-1==u&&c--,"function"!=typeof i.type&&(i.__u|=65536)):u!==r&&(u==r-1?c--:u==r+1?c++:(u>r?c--:c++,i.__u|=65536))):i=t.__k[_]=null;if(f)for(_=0;_(null!=l&&0==(131072&l.__u)?1:0))for(;r>=0||u=0){if((l=n[r])&&0==(131072&l.__u)&&i==l.key&&o===l.type)return r;r--}if(u2&&(u.children=arguments.length>3?w.call(arguments,2):e),I(t.type,u,_||t.key,i||t.ref,null)}function ht(t,n){var e={__c:n="__cC"+D++,__:t,Consumer:function(t,n){return t.children(n)},Provider:function(t){var e,_;return this.getChildContext||(e=new Set,(_={})[n]=this,this.getChildContext=function(){return _},this.componentWillUnmount=function(){e=null},this.shouldComponentUpdate=function(t){this.props.value!==t.value&&e.forEach((function(t){t.__e=!0,G(t)}))},this.sub=function(t){e.add(t);var n=t.componentWillUnmount;t.componentWillUnmount=function(){e&&e.delete(t),n&&n.call(t)}}),t.children}};return e.Provider.__=e.Consumer.contextType=e}w=A.slice,S={__e:function(t,n,e,_){for(var i,o,r;n=n.__;)if((i=n.__c)&&!i.__)try{if((o=i.constructor)&&null!=o.getDerivedStateFromError&&(i.setState(o.getDerivedStateFromError(t)),r=i.__d),null!=i.componentDidCatch&&(i.componentDidCatch(t,_||{}),r=i.__d),r)return i.__E=i}catch(n){t=n}throw t}},x=0,C=function(t){return null!=t&&null==t.constructor},q.prototype.setState=function(t,n){var e;e=null!=this.__s&&this.__s!==this.state?this.__s:this.__s=L({},this.state),"function"==typeof t&&(t=t(L({},e),this.props)),t&&L(e,t),null!=t&&this.__v&&(n&&this._sb.push(n),G(this))},q.prototype.forceUpdate=function(t){this.__v&&(this.__e=!0,t&&this.__h.push(t),G(this))},q.prototype.render=j,U=[],H="function"==typeof Promise?Promise.prototype.then.bind(Promise.resolve()):setTimeout,P=function(t,n){return t.__v.__b-n.__v.__b},J.__r=0,N=0,$=et(!1),T=et(!0),D=0;var at,pt,dt,vt,yt=0,mt=[],gt=S,bt=gt.__b,kt=gt.__r,wt=gt.diffed,St=gt.__c,xt=gt.unmount,Ct=gt.__;function Ut(t,n){gt.__h&>.__h(pt,t,yt||n),yt=0;var e=pt.__H||(pt.__H={__:[],__h:[]});return t>=e.__.length&&e.__.push({}),e.__[t]}function Et(t){return yt=1,Ht(Bt,t)}function Ht(t,n,e){var _=Ut(at++,2);if(_.t=t,!_.__c&&(_.__=[e?e(n):Bt(void 0,n),function(t){var n=_.__N?_.__N[0]:_.__[0],e=_.t(n,t);n!==e&&(_.__N=[e,_.__[1]],_.__c.setState({}))}],_.__c=pt,!pt.u)){var i=function(t,n,e){if(!_.__c.__H)return!0;var i=_.__c.__H.__.filter((function(t){return!!t.__c}));if(i.every((function(t){return!t.__N})))return!o||o.call(this,t,n,e);var r=!1;return i.forEach((function(t){if(t.__N){var n=t.__[0];t.__=t.__N,t.__N=void 0,n!==t.__[0]&&(r=!0)}})),!(!r&&_.__c.props===t)&&(!o||o.call(this,t,n,e))};pt.u=!0;var o=pt.shouldComponentUpdate,r=pt.componentWillUpdate;pt.componentWillUpdate=function(t,n,e){if(this.__e){var _=o;o=void 0,i(t,n,e),o=_}r&&r.call(this,t,n,e)},pt.shouldComponentUpdate=i}return _.__N||_.__}function Pt(t,n){var e=Ut(at++,3);!gt.__s&&qt(e.__H,n)&&(e.__=t,e.i=n,pt.__H.__h.push(e))}function Nt(t,n){var e=Ut(at++,4);!gt.__s&&qt(e.__H,n)&&(e.__=t,e.i=n,pt.__h.push(e))}function $t(t){return yt=5,Dt((function(){return{current:t}}),[])}function Tt(t,n,e){yt=6,Nt((function(){return"function"==typeof t?(t(n()),function(){return t(null)}):t?(t.current=n(),function(){return t.current=null}):void 0}),null==e?e:e.concat(t))}function Dt(t,n){var e=Ut(at++,7);return qt(e.__H,n)&&(e.__=t(),e.__H=n,e.__h=t),e.__}function Mt(t,n){return yt=8,Dt((function(){return t}),n)}function At(t){var n=pt.context[t.__c],e=Ut(at++,9);return e.c=t,n?(null==e.__&&(e.__=!0,n.sub(pt)),n.props.value):t.__}function Ft(t,n){gt.useDebugValue&>.useDebugValue(n?n(t):t)}function Wt(t){var n=Ut(at++,10),e=Et();return n.__=t,pt.componentDidCatch||(pt.componentDidCatch=function(t,_){n.__&&n.__(t,_),e[1](t)}),[e[0],function(){e[1](void 0)}]}function Lt(){var t=Ut(at++,11);if(!t.__){for(var n=pt.__v;null!==n&&!n.__m&&null!==n.__;)n=n.__;var e=n.__m||(n.__m=[0,0]);t.__="P"+e[0]+"-"+e[1]++}return t.__}function Ot(){for(var t;t=mt.shift();)if(t.__P&&t.__H)try{t.__H.__h.forEach(Vt),t.__H.__h.forEach(jt),t.__H.__h=[]}catch(n){t.__H.__h=[],gt.__e(n,t.__v)}}gt.__b=function(t){pt=null,bt&&bt(t)},gt.__=function(t,n){t&&n.__k&&n.__k.__m&&(t.__m=n.__k.__m),Ct&&Ct(t,n)},gt.__r=function(t){kt&&kt(t),at=0;var n=(pt=t.__c).__H;n&&(dt===pt?(n.__h=[],pt.__h=[],n.__.forEach((function(t){t.__N&&(t.__=t.__N),t.i=t.__N=void 0}))):(n.__h.forEach(Vt),n.__h.forEach(jt),n.__h=[],at=0)),dt=pt},gt.diffed=function(t){wt&&wt(t);var n=t.__c;n&&n.__H&&(n.__H.__h.length&&(1!==mt.push(n)&&vt===gt.requestAnimationFrame||((vt=gt.requestAnimationFrame)||It)(Ot)),n.__H.__.forEach((function(t){t.i&&(t.__H=t.i),t.i=void 0}))),dt=pt=null},gt.__c=function(t,n){n.some((function(t){try{t.__h.forEach(Vt),t.__h=t.__h.filter((function(t){return!t.__||jt(t)}))}catch(r){n.some((function(t){t.__h&&(t.__h=[])})),n=[],gt.__e(r,t.__v)}})),St&&St(t,n)},gt.unmount=function(t){xt&&xt(t);var n,e=t.__c;e&&e.__H&&(e.__H.__.forEach((function(t){try{Vt(t)}catch(t){n=t}})),e.__H=void 0,n&>.__e(n,e.__v))};var Rt="function"==typeof requestAnimationFrame;function It(t){var n,e=function(){clearTimeout(_),Rt&&cancelAnimationFrame(n),setTimeout(t)},_=setTimeout(e,100);Rt&&(n=requestAnimationFrame(e))}function Vt(t){var n=pt,e=t.__c;"function"==typeof e&&(t.__c=void 0,e()),pt=n}function jt(t){var n=pt;t.__c=t.__(),pt=n}function qt(t,n){return!t||t.length!==n.length||n.some((function(n,e){return n!==t[e]}))}function Bt(t,n){return"function"==typeof n?n(t):n}function zt(t,n){S[t]=n.bind(null,S[t]||(()=>{}))}let Gt,Jt;function Kt(t){if(Jt)Jt();Jt=t&&t.S()}function Qt({data:t}){const n=Yt(t);n.value=t;const e=Dt(()=>{let t=this.__v;while(t=t.__)if(t.__c){t.__c.__$f|=4;break}this.__$u.c=()=>{var t;if(!C(e.peek())&&3===(null==(t=this.base)?void 0:t.nodeType))this.base.data=e.peek();else{this.__$f|=1;this.setState({})}};return v(()=>{let t=n.value.value;return 0===t?0:!0===t?"":t||""})},[]);return e.value}Qt.displayName="_st";Object.defineProperties(f.prototype,{constructor:{configurable:!0,value:void 0},type:{configurable:!0,value:Qt},props:{configurable:!0,get(){return{data:this}}},__b:{configurable:!0,value:1}});zt("__b",(t,n)=>{if("string"==typeof n.type){let t,e=n.props;for(let _ in e){if("children"===_)continue;let i=e[_];if(i instanceof f){if(!t)n.__np=t={};t[_]=i;e[_]=i.peek()}}}t(n)});zt("__r",(t,n)=>{Kt();let e,_=n.__c;if(_){_.__$f&=-2;e=_.__$u;if(void 0===e)_.__$u=e=function(t){let n;k((function(){n=this}));n.c=()=>{_.__$f|=1;_.setState({})};return n}()}Gt=_;Kt(e);t(n)});zt("__e",(t,n,e,_)=>{Kt();Gt=void 0;t(n,e,_)});zt("diffed",(t,n)=>{Kt();Gt=void 0;let e;if("string"==typeof n.type&&(e=n.__e)){let t=n.__np,_=n.props;if(t){let n=e.U;if(n)for(let e in n){let _=n[e];if(void 0!==_&&!(e in t)){_.d();n[e]=void 0}}else{n={};e.U=n}for(let i in t){let o=n[i],r=t[i];if(void 0===o){o=Xt(e,i,r,_);n[i]=o}else o.o(r,_)}}}t(n)});function Xt(t,n,e,_){const i=n in t&&void 0===t.ownerSVGElement,o=c(e);return{o:(t,n)=>{o.value=t;_=n},d:k(()=>{const e=o.value.value;if(_[n]!==e){_[n]=e;if(i)t[n]=e;else if(e)t.setAttribute(n,e);else t.removeAttribute(n)}})}}zt("unmount",(t,n)=>{if("string"==typeof n.type){let t=n.__e;if(t){const n=t.U;if(n){t.U=void 0;for(let t in n){let e=n[t];if(e)e.d()}}}}else{let t=n.__c;if(t){const n=t.__$u;if(n){t.__$u=void 0;n.d()}}}t(n)});zt("__h",(t,n,e,_)=>{if(_<3||9===_)n.__$f|=2;t(n,e,_)});q.prototype.shouldComponentUpdate=function(t,n){const e=this.__$u;if(!(e&&void 0!==e.s||4&this.__$f))return!0;if(3&this.__$f)return!0;for(let _ in n)return!0;for(let _ in t)if("__source"!==_&&t[_]!==this.props[_])return!0;for(let _ in this.props)if(!(_ in t))return!0;return!1};function Yt(t){return Dt(()=>c(t),[])}function Zt(t){const n=$t(t);n.current=t;Gt.__$f|=4;return Dt(()=>v(()=>n.current()),[])}function tn(t){const n=$t(t);n.current=t;Pt(()=>k(()=>n.current()),[])}var nn=function(t,n,e,_){var i;n[0]=0;for(var o=1;o=5&&((i||!t&&5===_)&&(r.push(_,0,i,e),_=6),t&&(r.push(_,t,0,e),_=6)),i=""},l=0;l"===n?(_=1,i=""):i=n+i[0]:o?n===o?o="":i+=n:'"'===n||"'"===n?o=n:">"===n?(u(),_=1):_&&("="===n?(_=5,e=i,i=""):"/"===n&&(_<5||">"===t[l][s+1])?(u(),3===_&&(r=r[0]),_=r,(r=r[0]).push(2,0,_),_=0):" "===n||"\t"===n||"\n"===n||"\r"===n?(u(),_=2):i+=n),3===_&&"!--"===i&&(_=4,r=r[0])}return u(),r}(t)),n),arguments,[])).length>1?n:n[0]}var on=_n.bind(R);export{q as Component,j as Fragment,f as Signal,e as batch,ct as cloneElement,v as computed,ht as createContext,R as createElement,V as createRef,k as effect,R as h,on as html,ft as hydrate,C as isValidElement,S as options,st as render,c as signal,Y as toChildArray,o as untracked,Mt as useCallback,Zt as useComputed,At as useContext,Ft as useDebugValue,Pt as useEffect,Wt as useErrorBoundary,Lt as useId,Tt as useImperativeHandle,Nt as useLayoutEffect,Dt as useMemo,Ht as useReducer,$t as useRef,Yt as useSignal,tn as useSignalEffect,Et as useState};

From fbc98b748e7b075e327bcf13237057f647678049 Mon Sep 17 00:00:00 2001
From: MaggotHATE 
Date: Tue, 15 Oct 2024 15:54:55 +0500
Subject: [PATCH 14/36] sampling : add XTC sampler (#9742)

* Initial XTC commit

Adds XTC sampler, not activated by default, but recommended settings by default.

* Cleanup

* Simplified chances calculation

To be more inline with the original implementation, chance is calculated once at the beginning.

* First fixes by comments

Still need to look into sorting

* Fixed trailing backspaces

* Fixed RNG to be reproduceable

Thanks to @slaren for directions

* Fixed forgotten header

* Moved `min_keep`

Moved from conditions to a simple check at the end.

* Fixed broken randomization

Thanks to @slaren for explanation

* Swapped sorting for a custom algorithm

Shifts tokens to remove the penalized ones, then puts the penalized at the back. Should make `min_keep` still viable.

* Algorithm rework

1. Scan token from top till the first non-penalizable
2. Remove the last captured token (the least probable above threshold)
3. Shift all tokens to override the remaining penalizable
4. Penalize and put them at the the bottom.

* Added XTC to `test-sampling`

* Simplified algorithm and more tests

* Updated info in common and args

* Merged back lost commits in common and arg

* Update dump info in common

* Fixed incorrect min_keep check

* Added XTC to README

* Renamed parameters, fixed info and defaults

* probability is at 0 by default, but XTC is included in sampling queue
* threshold higher than 0.5 switches XTC off

* Initial server support

* Added XTC to server UIs

* Fixed labels in old server UI

* Made algorithm safer and more readable

* Removed xtc_threshold_max

* Fixed arg after update

* Quick fixes by comments

* Simplified algorithm since threshold_max is removed

* Renamed random distribution

* Fixed tests and outdated README

* Small fixes
---
 common/arg.cpp                        | 14 ++++
 common/common.cpp                     |  2 +
 common/common.h                       |  6 ++
 common/sampling.cpp                   | 13 +++-
 examples/main/README.md               | 13 ++++
 examples/server/public/index-new.html |  6 ++
 examples/server/public/index.html     |  4 ++
 examples/server/server.cpp            |  4 ++
 include/llama.h                       |  3 +
 src/llama-sampling.cpp                | 95 +++++++++++++++++++++++++++
 tests/test-sampling.cpp               | 45 +++++++++++--
 11 files changed, 195 insertions(+), 10 deletions(-)

diff --git a/common/arg.cpp b/common/arg.cpp
index 8969fc107..d6a8e1f6f 100644
--- a/common/arg.cpp
+++ b/common/arg.cpp
@@ -947,6 +947,20 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.sparams.tfs_z = std::stof(value);
         }
     ).set_sparam());
+    add_opt(common_arg(
+        {"--xtc-probability"}, "N",
+        string_format("xtc probability (default: %.1f, 0.0 = disabled)", (double)params.sparams.xtc_probability),
+        [](common_params & params, const std::string & value) {
+            params.sparams.xtc_probability = std::stof(value);
+        }
+    ).set_sparam());
+    add_opt(common_arg(
+        {"--xtc-threshold"}, "N",
+        string_format("xtc threshold (default: %.1f, 1.0 = disabled)", (double)params.sparams.xtc_threshold),
+        [](common_params & params, const std::string & value) {
+            params.sparams.xtc_threshold = std::stof(value);
+        }
+    ).set_sparam());
     add_opt(common_arg(
         {"--typical"}, "N",
         string_format("locally typical sampling, parameter p (default: %.1f, 1.0 = disabled)", (double)params.sparams.typ_p),
diff --git a/common/common.cpp b/common/common.cpp
index 451307b55..c08f01b42 100644
--- a/common/common.cpp
+++ b/common/common.cpp
@@ -2104,6 +2104,8 @@ void yaml_dump_non_result_info(FILE * stream, const common_params & params, cons
     fprintf(stream, "top_k: %d # default: 40\n", sparams.top_k);
     fprintf(stream, "top_p: %f # default: 0.95\n", sparams.top_p);
     fprintf(stream, "min_p: %f # default: 0.0\n", sparams.min_p);
+    fprintf(stream, "xtc_probability: %f # default: 0.0\n", sparams.xtc_probability);
+    fprintf(stream, "xtc_threshold: %f # default: 0.1\n", sparams.xtc_threshold);
     fprintf(stream, "typ_p: %f # default: 1.0\n", sparams.typ_p);
     fprintf(stream, "verbose_prompt: %s # default: false\n", params.verbose_prompt ? "true" : "false");
     fprintf(stream, "display_prompt: %s # default: true\n", params.display_prompt ? "true" : "false");
diff --git a/common/common.h b/common/common.h
index 5507b1c59..df2ee6bd4 100644
--- a/common/common.h
+++ b/common/common.h
@@ -90,6 +90,8 @@ enum common_sampler_type {
     COMMON_SAMPLER_TYPE_TFS_Z       = 4,
     COMMON_SAMPLER_TYPE_TYPICAL_P   = 5,
     COMMON_SAMPLER_TYPE_TEMPERATURE = 6,
+    COMMON_SAMPLER_TYPE_XTC         = 7,
+
 };
 
 // dimensionality reduction methods, used by cvector-generator
@@ -108,6 +110,8 @@ struct common_sampler_params {
     int32_t top_k             = 40;    // <= 0 to use vocab size
     float   top_p             = 0.95f; // 1.0 = disabled
     float   min_p             = 0.05f; // 0.0 = disabled
+    float   xtc_probability   = 0.00f; // 0.0 = disabled
+    float   xtc_threshold     = 0.10f; // > 0.5 disables XTC
     float   tfs_z             = 1.00f; // 1.0 = disabled
     float   typ_p             = 1.00f; // typical_p, 1.0 = disabled
     float   temp              = 0.80f; // <= 0.0 to sample greedily, 0.0 to not output probabilities
@@ -124,12 +128,14 @@ struct common_sampler_params {
     bool    ignore_eos        = false;
     bool    no_perf           = false; // disable performance metrics
 
+
     std::vector samplers = {
         COMMON_SAMPLER_TYPE_TOP_K,
         COMMON_SAMPLER_TYPE_TFS_Z,
         COMMON_SAMPLER_TYPE_TYPICAL_P,
         COMMON_SAMPLER_TYPE_TOP_P,
         COMMON_SAMPLER_TYPE_MIN_P,
+        COMMON_SAMPLER_TYPE_XTC,
         COMMON_SAMPLER_TYPE_TEMPERATURE
     };
 
diff --git a/common/sampling.cpp b/common/sampling.cpp
index cd49ade69..fb95bcd3b 100644
--- a/common/sampling.cpp
+++ b/common/sampling.cpp
@@ -130,10 +130,10 @@ std::string common_sampler_params::print() const {
 
     snprintf(result, sizeof(result),
             "\trepeat_last_n = %d, repeat_penalty = %.3f, frequency_penalty = %.3f, presence_penalty = %.3f\n"
-            "\ttop_k = %d, tfs_z = %.3f, top_p = %.3f, min_p = %.3f, typical_p = %.3f, temp = %.3f\n"
+            "\ttop_k = %d, tfs_z = %.3f, top_p = %.3f, min_p = %.3f, xtc_probability = %.3f, xtc_threshold = %.3f, typical_p = %.3f, temp = %.3f\n"
             "\tmirostat = %d, mirostat_lr = %.3f, mirostat_ent = %.3f",
             penalty_last_n, penalty_repeat, penalty_freq, penalty_present,
-            top_k, tfs_z, top_p, min_p, typ_p, temp,
+            top_k, tfs_z, top_p, min_p, xtc_probability, xtc_threshold, typ_p, temp,
             mirostat, mirostat_eta, mirostat_tau);
 
     return std::string(result);
@@ -184,6 +184,9 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
                     case COMMON_SAMPLER_TYPE_MIN_P:
                         llama_sampler_chain_add(result->chain, llama_sampler_init_min_p    (params.min_p, params.min_keep));
                         break;
+                    case COMMON_SAMPLER_TYPE_XTC:
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_xtc      (params.xtc_probability, params.xtc_threshold, params.min_keep, params.seed));
+                        break;
                     case COMMON_SAMPLER_TYPE_TFS_Z:
                         llama_sampler_chain_add(result->chain, llama_sampler_init_tail_free(params.tfs_z, params.min_keep));
                         break;
@@ -372,6 +375,7 @@ char common_sampler_type_to_chr(enum common_sampler_type cnstr) {
         case COMMON_SAMPLER_TYPE_TOP_P:       return 'p';
         case COMMON_SAMPLER_TYPE_MIN_P:       return 'm';
         case COMMON_SAMPLER_TYPE_TEMPERATURE: return 't';
+        case COMMON_SAMPLER_TYPE_XTC:         return 'x';
         default : return '?';
     }
 }
@@ -384,6 +388,7 @@ std::string common_sampler_type_to_str(enum common_sampler_type cnstr) {
         case COMMON_SAMPLER_TYPE_TOP_P:       return "top_p";
         case COMMON_SAMPLER_TYPE_MIN_P:       return "min_p";
         case COMMON_SAMPLER_TYPE_TEMPERATURE: return "temperature";
+        case COMMON_SAMPLER_TYPE_XTC:         return "xtc";
         default : return "";
     }
 }
@@ -396,6 +401,7 @@ std::vector common_sampler_types_from_names(const std::vect
         { "min_p",       COMMON_SAMPLER_TYPE_MIN_P },
         { "tfs_z",       COMMON_SAMPLER_TYPE_TFS_Z },
         { "temperature", COMMON_SAMPLER_TYPE_TEMPERATURE },
+        { "xtc",         COMMON_SAMPLER_TYPE_XTC },
     };
 
     // since samplers names are written multiple ways
@@ -441,7 +447,8 @@ std::vector common_sampler_types_from_chars(const std::stri
         { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_TYPICAL_P),   COMMON_SAMPLER_TYPE_TYPICAL_P },
         { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_TOP_P),       COMMON_SAMPLER_TYPE_TOP_P },
         { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_MIN_P),       COMMON_SAMPLER_TYPE_MIN_P },
-        { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_TEMPERATURE), COMMON_SAMPLER_TYPE_TEMPERATURE }
+        { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_TEMPERATURE), COMMON_SAMPLER_TYPE_TEMPERATURE },
+        { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_XTC),         COMMON_SAMPLER_TYPE_XTC }
     };
 
     std::vector samplers;
diff --git a/examples/main/README.md b/examples/main/README.md
index f0c3031ab..620934dad 100644
--- a/examples/main/README.md
+++ b/examples/main/README.md
@@ -241,6 +241,19 @@ The `--mirostat-ent` option sets the Mirostat target entropy (tau), which repres
 
 Example usage: `--mirostat 2 --mirostat-lr 0.05 --mirostat-ent 3.0`
 
+### XTC Sampling
+
+-   `--xtc-probability N`: Sets the chance for token removal (checked once on sampler start) (default: 0.0).
+-   `--xtc-threshold N`: Sets a minimum probability threshold for tokens to be removed (default: 0.1).
+
+Exclude Top Choices (XTC) is a unique sampler that is designed to remove top tokens from consideration and avoid more obvious and repetitive outputs. With a chance of `xtc-probability` it searches for tokens with probabilities of `xtc-threshold` and above, then removes all such tokens except the least probable one.
+
+By removing top tokens XTC can improve the variety of answers, break writing clichés and inhibit repition, since clichés and repeated phrases are usually more likely to appear. By keeping the last token above the threshold, XTC ensures that the answer is still coherent. XTC is meant to be used for creative tasks, but feel free to experiment with different settings for different models.
+
+Being experimental and unique, XTC is disabled by default. The recommended combination of samplers is Min-P followed by XTC on its default settings: `--sampling-seq mx --min-p 0.02 --xtc-probability 0.5`.
+
+Example usage: `--xtc-probability 0.5 --xtc-threshold 0.1`
+
 ### Logit Bias
 
 -   `-l TOKEN_ID(+/-)BIAS, --logit-bias TOKEN_ID(+/-)BIAS`: Modify the likelihood of a token appearing in the generated text completion.
diff --git a/examples/server/public/index-new.html b/examples/server/public/index-new.html
index c87dd8f1e..ad4183cd9 100644
--- a/examples/server/public/index-new.html
+++ b/examples/server/public/index-new.html
@@ -43,6 +43,8 @@
       top_k: 0, // <= 0 to use vocab size
       top_p: 1.0, // 1.0 = disabled
       min_p: 0.05, // 0 = disabled; recommended for non-english: ~ 0.4
+      xtc_probability: 0.0, // 0 = disabled;
+      xtc_threshold: 0.1, // > 0.5 disables XTC;
       tfs_z: 1.0, // 1.0 = disabled
       typical_p: 1.0, // 1.0 = disabled
       presence_penalty: 0.0, // 0.0 = disabled
@@ -836,6 +838,8 @@ return html`
           ${FloatField({ label: "TFS-Z", title: "Activates tail-free sampling, a method used to limit the prediction of tokens that are too frequent. The parameter z controls the strength of this limitation. A value of 1.0 means that this function is deactivated.", max: 1.0, min: 0.0, name: "tfs_z", step: 0.01, value: params.value.tfs_z })}
           ${FloatField({ label: "Frequency Penalty", title: "A penalty that is applied based on the frequency with which certain tokens occur in the training data set. A higher value results in rare tokens being favoured.", max: 1.0, min: 0.0, name: "frequency_penalty", step: 0.01, value: params.value.frequency_penalty })}
           ${FloatField({ label: "Typical-P", title: "Activates local typical sampling, a method used to limit the prediction of tokens that are atypical in the current context. The parameter p controls the strength of this limitation. A value of 1.0 means that this function is deactivated.", max: 1.0, min: 0.0, name: "typical_p", step: 0.01, value: params.value.typical_p })}
+          ${FloatField({ label: "XTC probability", title: "Sets the chance for token removal (checked once on sampler start)", max: 1.0, min: 0.0, name: "xtc_probability", step: 0.01, value: params.value.xtc_probability })}
+          ${FloatField({ label: "XTC threshold", title: "Sets a minimum probability threshold for tokens to be removed", max: 0.5, min: 0.0, name: "xtc_threshold", step: 0.01, value: params.value.xtc_threshold })}
           ${IntField({ label: "Min Keep", title: "If greater than 0, samplers are forced to return N possible tokens at minimum. Default is 0", max: 10, min: 0, name: "min_keep", value: params.value.min_keep })}
         
 
@@ -1132,6 +1136,8 @@ document.addEventListener('DOMContentLoaded', (event) => {
   const snapSettings = {
     temperature: { snapValue: 1.0, snapRangeMultiplier: 6 },
     min_p: { snapValue: 0.05, snapRangeMultiplier: 2 },
+    xtc_probability: { snapValue: 0.0, snapRangeMultiplier: 4 },
+    xtc_threshold: { snapValue: 0.5, snapRangeMultiplier: 4 },
     top_p: { snapValue: 1.0, snapRangeMultiplier: 4 },
     tfs_z: { snapValue: 1.0, snapRangeMultiplier: 4 },
     typical_p: { snapValue: 1.0, snapRangeMultiplier: 4 },
diff --git a/examples/server/public/index.html b/examples/server/public/index.html
index 07fec6a38..88065705f 100644
--- a/examples/server/public/index.html
+++ b/examples/server/public/index.html
@@ -307,6 +307,8 @@
       top_k: 40, // <= 0 to use vocab size
       top_p: 0.95, // 1.0 = disabled
       min_p: 0.05, // 0 = disabled
+      xtc_probability: 0.0, // 0 = disabled;
+      xtc_threshold: 0.1, // > 0.5 disables XTC;
       tfs_z: 1.0, // 1.0 = disabled
       typical_p: 1.0, // 1.0 = disabled
       presence_penalty: 0.0, // 0.0 = disabled
@@ -1013,6 +1015,8 @@
               ${FloatField({ label: "Typical P", max: 1.0, min: 0.0, name: "typical_p", step: 0.01, value: params.value.typical_p })}
               ${FloatField({ label: "Presence penalty", max: 1.0, min: 0.0, name: "presence_penalty", step: 0.01, value: params.value.presence_penalty })}
               ${FloatField({ label: "Frequency penalty", max: 1.0, min: 0.0, name: "frequency_penalty", step: 0.01, value: params.value.frequency_penalty })}
+              ${FloatField({ label: "XTC probability", max: 1.0, min: 0.0, name: "xtc_probability", step: 0.01, value: params.value.xtc_probability })}
+              ${FloatField({ label: "XTC threshold", max: 0.5, min: 0.0, name: "xtc_threshold", step: 0.01, value: params.value.xtc_threshold })}
             
             

             

diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index 18bcad3f0..8d4380e12 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -863,6 +863,8 @@ struct server_context {
         slot.sparams.top_k             = json_value(data, "top_k",             default_sparams.top_k);
         slot.sparams.top_p             = json_value(data, "top_p",             default_sparams.top_p);
         slot.sparams.min_p             = json_value(data, "min_p",             default_sparams.min_p);
+        slot.sparams.xtc_probability   = json_value(data, "xtc_probability",   default_sparams.xtc_probability);
+        slot.sparams.xtc_threshold     = json_value(data, "xtc_threshold",     default_sparams.xtc_threshold);
         slot.sparams.tfs_z             = json_value(data, "tfs_z",             default_sparams.tfs_z);
         slot.sparams.typ_p             = json_value(data, "typical_p",         default_sparams.typ_p);
         slot.sparams.temp              = json_value(data, "temperature",       default_sparams.temp);
@@ -1196,6 +1198,8 @@ struct server_context {
             {"top_k",                     slot.sparams.top_k},
             {"top_p",                     slot.sparams.top_p},
             {"min_p",                     slot.sparams.min_p},
+            {"xtc_probability",           slot.sparams.xtc_probability},
+            {"xtc_threshold",             slot.sparams.xtc_threshold},
             {"tfs_z",                     slot.sparams.tfs_z},
             {"typical_p",                 slot.sparams.typ_p},
             {"repeat_last_n",             slot.sparams.penalty_last_n},
diff --git a/include/llama.h b/include/llama.h
index 9110b5956..92d4c70c1 100644
--- a/include/llama.h
+++ b/include/llama.h
@@ -1101,6 +1101,9 @@ extern "C" {
     /// @details Dynamic temperature implementation (a.k.a. entropy) described in the paper https://arxiv.org/abs/2309.02772.
     LLAMA_API struct llama_sampler * llama_sampler_init_temp_ext   (float   t, float   delta, float exponent);
 
+    /// @details XTC sampler as described in https://github.com/oobabooga/text-generation-webui/pull/6335
+    LLAMA_API struct llama_sampler * llama_sampler_init_xtc        (float   p, float   t,     size_t min_keep, uint32_t seed);
+
     /// @details Mirostat 1.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
     /// @param candidates A vector of `llama_token_data` containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text.
     /// @param tau  The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
diff --git a/src/llama-sampling.cpp b/src/llama-sampling.cpp
index e255a8fc4..67a78c3ac 100644
--- a/src/llama-sampling.cpp
+++ b/src/llama-sampling.cpp
@@ -1059,6 +1059,101 @@ struct llama_sampler * llama_sampler_init_temp_ext(float temp, float delta, floa
     };
 }
 
+// xtc
+
+struct llama_sampler_xtc {
+    const float    probability;
+    const float    threshold;
+    const size_t   min_keep;
+
+    const uint32_t seed;
+    uint32_t       seed_cur;
+
+    std::mt19937   rng;
+};
+
+static const char * llama_sampler_xtc_name(const struct llama_sampler * /*smpl*/) {
+    return "xtc";
+}
+
+static void llama_sample_xtc_apply(struct llama_sampler * smpl, llama_token_data_array * cur_p) {
+    auto * ctx = (llama_sampler_xtc *) smpl->ctx;
+
+    if (ctx->probability <= 0.0f
+        || ctx->threshold > 0.5f
+        || cur_p->size < 2) {
+        return;
+    }
+
+    std::uniform_real_distribution distribution(0.0f, 1.0f);
+    float chance = distribution(ctx->rng);
+    if (chance > ctx->probability) return;
+
+    // in case it's not sorted/recalculated yet
+    llama_sampler_softmax_impl(cur_p);
+
+    int pos_last = 0;
+
+    for (size_t i = 0; i < cur_p->size; ++i) {
+        if (cur_p->data[i].p >= ctx->threshold) {
+            pos_last = i;
+        } else break;
+    }
+
+    if (cur_p->size - pos_last >= ctx->min_keep && pos_last > 0) {
+        cur_p->data += pos_last;
+        cur_p->size -= pos_last;
+    }
+}
+
+static struct llama_sampler * llama_sampler_xtc_clone(const struct llama_sampler * smpl) {
+    const auto * ctx = (const llama_sampler_xtc *) smpl->ctx;
+    auto * result = llama_sampler_init_xtc(ctx->probability, ctx->threshold, ctx->min_keep, ctx->seed);
+
+    // copy the state
+    {
+        auto * result_ctx = (llama_sampler_xtc *) result->ctx;
+
+        result_ctx->rng = ctx->rng;
+    }
+
+    return result;
+}
+
+static void llama_sampler_xtc_free(struct llama_sampler * smpl) {
+    delete (llama_sampler_xtc *) smpl->ctx;
+}
+
+static void llama_sampler_xtc_reset(struct llama_sampler * smpl) {
+    auto * ctx = (llama_sampler_xtc *) smpl->ctx;
+    ctx->seed_cur = get_rng_seed(ctx->seed);
+    ctx->rng.seed(ctx->seed_cur);
+}
+
+static struct llama_sampler_i llama_sampler_xtc_i = {
+    /* .name   = */ llama_sampler_xtc_name,
+    /* .accept = */ nullptr,
+    /* .apply  = */ llama_sample_xtc_apply,
+    /* .reset  = */ llama_sampler_xtc_reset,
+    /* .clone  = */ llama_sampler_xtc_clone,
+    /* .free   = */ llama_sampler_xtc_free,
+};
+
+struct llama_sampler * llama_sampler_init_xtc(float p, float t, size_t min_keep, uint32_t seed) {
+    auto seed_cur = get_rng_seed(seed);
+    return new llama_sampler {
+        /* .iface = */ &llama_sampler_xtc_i,
+        /* .ctx   = */ new llama_sampler_xtc {
+            /* .probability   = */ p,
+            /* .threshold     = */ t,
+            /* .min_keep      = */ min_keep,
+            /* .seed          = */ seed,
+            /* .seed_cur      = */ seed_cur,
+            /* .rng           = */ std::mt19937(seed_cur),
+        },
+    };
+}
+
 // mirostat
 
 struct llama_sampler_mirostat {
diff --git a/tests/test-sampling.cpp b/tests/test-sampling.cpp
index 6e021c4c7..1372bdf13 100644
--- a/tests/test-sampling.cpp
+++ b/tests/test-sampling.cpp
@@ -111,6 +111,28 @@ static void test_min_p(const std::vector & probs, const std::vector & probs, const std::vector & expected_probs, float p, float t) {
+    const size_t n_vocab = probs.size();
+
+    std::vector cur;
+    cur.reserve(n_vocab);
+    for (llama_token token_id = 0; token_id < (llama_token)n_vocab; token_id++) {
+        const float logit = logf(probs[token_id]);
+        cur.emplace_back(llama_token_data{token_id, logit, 0.0f});
+    }
+
+    llama_token_data_array cur_p = { cur.data(), cur.size(), -1, false };
+    APPLY(llama_sampler_init_softmax(), &cur_p);
+    DUMP(&cur_p);
+    APPLY(llama_sampler_init_xtc(p, t, 0, 0), &cur_p);
+    DUMP(&cur_p);
+
+    GGML_ASSERT(cur_p.size == expected_probs.size());
+    for (size_t i = 0; i < cur_p.size; i++) {
+        GGML_ASSERT(fabs(cur_p.data[i].p - expected_probs[i]) < 1e-5);
+    }
+}
+
 static void test_typical(const std::vector & probs, const std::vector & expected_probs, float p) {
     const size_t n_vocab = probs.size();
 
@@ -263,7 +285,7 @@ static void bench(llama_sampler * cnstr, const char * cnstr_name, const std::vec
     }
     const int64_t t_end = ggml_time_us();
     llama_sampler_free(cnstr);
-    printf("%-42s: %8.3f us/iter\n", cnstr_name, (t_end - t_start) / (float)n_iter);
+    printf("%-43s: %8.3f us/iter\n", cnstr_name, (t_end - t_start) / (float)n_iter);
 }
 
 #define BENCH(__cnstr, __data, __n_iter) bench((__cnstr), #__cnstr, (__data), (__n_iter))
@@ -279,12 +301,13 @@ static void test_perf() {
         data.emplace_back(llama_token_data{i, logit, 0.0f});
     }
 
-    BENCH(llama_sampler_init_top_k    (40),      data, 32);
-    BENCH(llama_sampler_init_top_p    (0.8f, 1), data, 32);
-    BENCH(llama_sampler_init_min_p    (0.2f, 1), data, 32);
-    BENCH(llama_sampler_init_tail_free(0.5f, 1), data, 32);
-    BENCH(llama_sampler_init_typical  (0.5f, 1), data, 32);
-    BENCH(llama_sampler_init_softmax  (),        data, 32);
+    BENCH(llama_sampler_init_top_k    (40),                     data, 32);
+    BENCH(llama_sampler_init_top_p    (0.8f, 1),                data, 32);
+    BENCH(llama_sampler_init_min_p    (0.2f, 1),                data, 32);
+    BENCH(llama_sampler_init_tail_free(0.5f, 1),                data, 32);
+    BENCH(llama_sampler_init_typical  (0.5f, 1),                data, 32);
+    BENCH(llama_sampler_init_xtc      (1.0f, 0.1f, 1, 1),       data, 32);
+    BENCH(llama_sampler_init_softmax  (),                       data, 32);
 }
 
 int main(void) {
@@ -309,6 +332,14 @@ int main(void) {
     test_min_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f/0.4f},                                  0.76f);
     test_min_p({0.1f, 0.2f, 0.3f, 0.4f}, {0.4f/0.4f},                                  1.00f);
 
+    printf("XTC should:\n");
+    test_xtc({0.4f, 0.3f, 0.2f, 0.1f},   {0.1f},                                0.99f, 0.09f);
+    test_xtc({0.4f, 0.3f, 0.2f, 0.1f},   {0.2f, 0.1f},                          0.99f, 0.19f);
+    test_xtc({0.4f, 0.3f, 0.2f, 0.1f},   {0.3f, 0.2f, 0.1f},                    0.99f, 0.29f);
+
+    printf("XTC should not:\n");
+    test_xtc({0.4f, 0.3f, 0.2f, 0.1f},   {0.4f, 0.3f, 0.2f, 0.1f},              0.99f, 0.39f);
+
     test_tfs({0.1f, 0.15f, 0.2f, 0.25f, 0.3f}, {0.3f}, 0.25f);
     test_tfs({0.1f, 0.15f, 0.2f, 0.25f, 0.3f}, {0.3f, 0.25f}, 0.75f);
     test_tfs({0.1f, 0.15f, 0.2f, 0.25f, 0.3f}, {0.3f, 0.25f}, 0.99f);

From 223c25a72fcc3f65cdfd7f5d57edd5b44b550e18 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Tue, 15 Oct 2024 16:28:55 +0300
Subject: [PATCH 15/36] server : improve infill context reuse (#9894)

ggml-ci
---
 examples/server/README.md  | 10 +++---
 examples/server/server.cpp | 73 ++++++++++++++------------------------
 2 files changed, 33 insertions(+), 50 deletions(-)

diff --git a/examples/server/README.md b/examples/server/README.md
index eb0a7b32e..fcdb02afd 100644
--- a/examples/server/README.md
+++ b/examples/server/README.md
@@ -524,10 +524,12 @@ Takes a prefix and a suffix and returns the predicted completion as stream.
 
 - `input_prefix`: Set the prefix of the code to infill.
 - `input_suffix`: Set the suffix of the code to infill.
-- `prompt`: Added after the `FIM_MID` token
-- `extra_context`: Additional context inserted before the FIM prefix. See https://github.com/ggerganov/llama.cpp/pull/9874
+- `input_extra`:  Additional context inserted before the FIM prefix.
+- `prompt`:       Added after the `FIM_MID` token
 
-It also accepts all the options of `/completion`.
+`input_extra` is array of `{"filename": string, "text": string}` objects.
+
+The endpoint also accepts all the options of `/completion`.
 
 If the model has `FIM_REPO` and `FIM_FILE_SEP` tokens, the [repo-level pattern](https://arxiv.org/pdf/2409.12186) is used:
 
@@ -545,7 +547,7 @@ If the model has `FIM_REPO` and `FIM_FILE_SEP` tokens, the [repo-level pattern](
 If the tokens are missing, then the extra context is simply prefixed at the start:
 
 ```txt
-[extra_context][input_prefix][input_suffix][prompt]
+[input_extra][input_prefix][input_suffix][prompt]
 ```
 
 ### **GET** `/props`: Get server global properties.
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index 8d4380e12..d53cca84c 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -136,10 +136,6 @@ struct slot_params {
     int64_t t_max_predict_ms = -1; // if positive, limit the generation phase to this time limit
 
     std::vector antiprompt;
-
-    json input_prefix;
-    json input_suffix;
-    json extra_context;
 };
 
 struct server_slot {
@@ -169,6 +165,10 @@ struct server_slot {
 
     json prompt; // can be either a string, array of strings or array of token ids
 
+    json input_prefix;
+    json input_suffix;
+    json input_extra;
+
     // when a task is submitted, we first tokenize the prompt and store it here
     std::vector prompt_tokens;
     std::vector extra_tokens;
@@ -910,12 +910,12 @@ struct server_context {
         }
 
         // infill
-        slot.params.input_prefix  = json_value(data, "input_prefix",  default_params.input_prefix);
-        slot.params.input_suffix  = json_value(data, "input_suffix",  default_params.input_suffix);
-        slot.params.extra_context = json_value(data, "extra_context", default_params.extra_context);
+        slot.input_prefix = json_value(data, "input_prefix", json());
+        slot.input_suffix = json_value(data, "input_suffix", json());
+        slot.input_extra  = json_value(data, "input_extra",  json());
 
-        SLT_DBG(slot, "extra_context chunks: %d\n", (int) slot.params.extra_context.size());
-        for (const auto & chunk : slot.params.extra_context) {
+        SLT_DBG(slot, "extra_context chunks: %d\n", (int) slot.input_extra.size());
+        for (const auto & chunk : slot.input_extra) {
             // { "text": string, "filename": string }
             if (!chunk.contains("text") || !chunk["text"].is_string()) {
                 send_error(task, "extra_context chunk must contain a \"text\" field with a string value", ERROR_TYPE_INVALID_REQUEST);
@@ -932,7 +932,7 @@ struct server_context {
         }
 
         // get prompt
-        if (task.cmpl_type != SERVER_TASK_CMPL_TYPE_INFILL) {
+        {
             const auto & prompt = data.find("prompt");
             if (prompt == data.end()) {
                 send_error(task, "\"prompt\" must be provided", ERROR_TYPE_INVALID_REQUEST);
@@ -1958,6 +1958,8 @@ struct server_context {
                                 } break;
                             case SERVER_TASK_CMPL_TYPE_INFILL:
                                 {
+                                    // TODO: optimize this block by reducing memory allocations and movement
+
                                     // use FIM repo-level pattern:
                                     // ref: https://arxiv.org/pdf/2409.12186
                                     //
@@ -1968,10 +1970,11 @@ struct server_context {
                                     // extra chunk 1
                                     // ...
                                     // [FIM_SEP]filename
-                                    // [FIM_PRE]prefix[FIM_SUF]suffix[FIM_MID]
+                                    // [FIM_PRE]prefix[FIM_SUF]suffix[FIM_MID]prompt
                                     //
-                                    auto prefix_tokens = tokenize(slot.params.input_prefix, false, false);
-                                    auto suffix_tokens = tokenize(slot.params.input_suffix, false, false);
+                                    auto tokens_prefix = tokenize(slot.input_prefix, false, false);
+                                    auto tokens_suffix = tokenize(slot.input_suffix, false, false);
+                                    auto tokens_prompt = tokenize(slot.prompt,       false, false);
 
                                     slot.extra_tokens.clear();
                                     if (llama_token_fim_rep(model) != LLAMA_TOKEN_NULL) {
@@ -1981,7 +1984,7 @@ struct server_context {
                                         slot.extra_tokens.insert(slot.extra_tokens.end(), k_fim_repo.begin(), k_fim_repo.end());
                                     }
 
-                                    for (const auto & chunk : slot.params.extra_context) {
+                                    for (const auto & chunk : slot.input_extra) {
                                         // { "text": string, "filename": string }
                                         const std::string text     = chunk.value("text", "");
                                         const std::string filename = chunk.value("filename", "tmp");
@@ -2012,20 +2015,21 @@ struct server_context {
                                     }
 
                                     // for now pick FIM context to fit in a batch (ratio prefix:suffix = 3:1, TODO: configurable?)
-                                    const int n_suffix_take = std::min(suffix_tokens.size(), (n_batch)/4);
-                                    const int n_prefix_take = std::min(prefix_tokens.size(), (n_batch - 3) - n_suffix_take);
+                                    const int n_suffix_take = std::min(tokens_suffix.size(),   (n_batch/4));
+                                    const int n_prefix_take = std::min(tokens_prefix.size(), 3*(n_batch/4) - 3);
 
                                     // fill the rest of the context with extra chunks
                                     const int n_extra_take = std::min(std::max(0, slot.n_ctx - (n_batch) - 2*slot.n_predict), slot.extra_tokens.size());
 
-                                    prefix_tokens.erase(prefix_tokens.begin(), prefix_tokens.begin() + prefix_tokens.size() - n_prefix_take);
-                                    suffix_tokens.resize(n_suffix_take);
+                                    tokens_prefix.erase(tokens_prefix.begin(), tokens_prefix.begin() + tokens_prefix.size() - n_prefix_take);
+                                    tokens_suffix.resize(n_suffix_take);
 
-                                    prefix_tokens.insert(prefix_tokens.begin(), llama_token_fim_pre(model));
-                                    suffix_tokens.insert(suffix_tokens.begin(), llama_token_fim_suf(model));
+                                    tokens_prefix.insert(tokens_prefix.begin(), llama_token_fim_pre(model));
+                                    tokens_prefix.insert(tokens_prefix.end(),   tokens_prompt.begin(), tokens_prompt.end());
+                                    tokens_suffix.insert(tokens_suffix.begin(), llama_token_fim_suf(model));
 
-                                    auto embd_inp = params.spm_infill ? suffix_tokens : prefix_tokens;
-                                    auto embd_end = params.spm_infill ? prefix_tokens : suffix_tokens;
+                                    auto embd_inp = params.spm_infill ? tokens_suffix : tokens_prefix;
+                                    auto embd_end = params.spm_infill ? tokens_prefix : tokens_suffix;
 
                                     if (llama_add_bos_token(model)) {
                                         embd_inp.insert(embd_inp.begin(), llama_token_bos(model));
@@ -2140,40 +2144,17 @@ struct server_context {
 
                                     while (head_c < slot.cache_tokens.size() &&
                                            head_p < prompt_tokens.size()) {
-                                        if (llama_token_is_control(model, slot.cache_tokens[head_c]) &&
-                                            slot.cache_tokens[head_c] != llama_token_fim_rep(model) &&
-                                            slot.cache_tokens[head_c] != llama_token_fim_sep(model)) {
-                                            break;
-                                        }
-
-                                        if (llama_token_is_control(model, prompt_tokens[head_p]) &&
-                                            prompt_tokens[head_p] != llama_token_fim_rep(model) &&
-                                            prompt_tokens[head_p] != llama_token_fim_sep(model)) {
-                                            break;
-                                        }
 
                                         size_t n_match = 0;
-
                                         while (head_c + n_match < slot.cache_tokens.size() &&
                                                head_p + n_match < prompt_tokens.size()     &&
                                                slot.cache_tokens[head_c + n_match] == prompt_tokens[head_p + n_match]) {
-                                            if (llama_token_is_control(model, slot.cache_tokens[head_c + n_match]) &&
-                                                slot.cache_tokens[head_c + n_match] != llama_token_fim_rep(model) &&
-                                                slot.cache_tokens[head_c + n_match] != llama_token_fim_sep(model)) {
-                                                break;
-                                            }
-
-                                            if (llama_token_is_control(model, prompt_tokens[head_p + n_match]) &&
-                                                prompt_tokens[head_p + n_match] != llama_token_fim_rep(model) &&
-                                                prompt_tokens[head_p + n_match] != llama_token_fim_sep(model)) {
-                                                break;
-                                            }
 
                                             n_match++;
                                         }
 
                                         if (n_match >= (size_t) params.n_cache_reuse) {
-                                            SLT_DBG(slot, "reusing chunk with size %zu, shifting KV cache [%zu, %zu) -> [%zu, %zu)\n", n_match, head_c, head_c + n_match, head_p, head_p + n_match);
+                                            SLT_INF(slot, "reusing chunk with size %zu, shifting KV cache [%zu, %zu) -> [%zu, %zu)\n", n_match, head_c, head_c + n_match, head_p, head_p + n_match);
                                             //for (size_t i = head_p; i < head_p + n_match; i++) {
                                             //    SLT_DBG(slot, "cache token %3zu: %6d '%s'\n", i, prompt_tokens[i], common_token_to_piece(ctx, prompt_tokens[i]).c_str());
                                             //}

From 755a9b2bf00fbae988e03a47e852b66eaddd113a Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Tue, 15 Oct 2024 16:35:33 +0300
Subject: [PATCH 16/36] llama : add infill sampler (#9896)

ggml-ci
---
 common/common.h        |   4 +-
 common/sampling.cpp    |   9 +-
 examples/main/main.cpp |  46 +++++-----
 include/llama.h        |  28 ++++++
 src/llama-sampling.cpp | 201 +++++++++++++++++++++++++++++++++++++++++
 src/llama-sampling.h   |   5 +-
 src/llama-vocab.cpp    |  17 ++++
 src/llama-vocab.h      |   8 +-
 src/llama.cpp          |  11 +++
 9 files changed, 300 insertions(+), 29 deletions(-)

diff --git a/common/common.h b/common/common.h
index df2ee6bd4..5ca8fd391 100644
--- a/common/common.h
+++ b/common/common.h
@@ -91,7 +91,7 @@ enum common_sampler_type {
     COMMON_SAMPLER_TYPE_TYPICAL_P   = 5,
     COMMON_SAMPLER_TYPE_TEMPERATURE = 6,
     COMMON_SAMPLER_TYPE_XTC         = 7,
-
+    COMMON_SAMPLER_TYPE_INFILL      = 8,
 };
 
 // dimensionality reduction methods, used by cvector-generator
@@ -136,7 +136,7 @@ struct common_sampler_params {
         COMMON_SAMPLER_TYPE_TOP_P,
         COMMON_SAMPLER_TYPE_MIN_P,
         COMMON_SAMPLER_TYPE_XTC,
-        COMMON_SAMPLER_TYPE_TEMPERATURE
+        COMMON_SAMPLER_TYPE_TEMPERATURE,
     };
 
     std::string grammar; // optional BNF-like grammar to constrain sampling
diff --git a/common/sampling.cpp b/common/sampling.cpp
index fb95bcd3b..56cd0df6b 100644
--- a/common/sampling.cpp
+++ b/common/sampling.cpp
@@ -196,6 +196,9 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
                     case COMMON_SAMPLER_TYPE_TEMPERATURE:
                         llama_sampler_chain_add(result->chain, llama_sampler_init_temp_ext (params.temp, params.dynatemp_range, params.dynatemp_exponent));
                         break;
+                    case COMMON_SAMPLER_TYPE_INFILL:
+                        llama_sampler_chain_add(result->chain, llama_sampler_init_infill   (model));
+                        break;
                     default:
                         GGML_ASSERT(false && "unknown sampler type");
                 }
@@ -376,6 +379,7 @@ char common_sampler_type_to_chr(enum common_sampler_type cnstr) {
         case COMMON_SAMPLER_TYPE_MIN_P:       return 'm';
         case COMMON_SAMPLER_TYPE_TEMPERATURE: return 't';
         case COMMON_SAMPLER_TYPE_XTC:         return 'x';
+        case COMMON_SAMPLER_TYPE_INFILL:      return 'i';
         default : return '?';
     }
 }
@@ -389,6 +393,7 @@ std::string common_sampler_type_to_str(enum common_sampler_type cnstr) {
         case COMMON_SAMPLER_TYPE_MIN_P:       return "min_p";
         case COMMON_SAMPLER_TYPE_TEMPERATURE: return "temperature";
         case COMMON_SAMPLER_TYPE_XTC:         return "xtc";
+        case COMMON_SAMPLER_TYPE_INFILL:      return "infill";
         default : return "";
     }
 }
@@ -402,6 +407,7 @@ std::vector common_sampler_types_from_names(const std::vect
         { "tfs_z",       COMMON_SAMPLER_TYPE_TFS_Z },
         { "temperature", COMMON_SAMPLER_TYPE_TEMPERATURE },
         { "xtc",         COMMON_SAMPLER_TYPE_XTC },
+        { "infill",      COMMON_SAMPLER_TYPE_INFILL },
     };
 
     // since samplers names are written multiple ways
@@ -448,7 +454,8 @@ std::vector common_sampler_types_from_chars(const std::stri
         { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_TOP_P),       COMMON_SAMPLER_TYPE_TOP_P },
         { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_MIN_P),       COMMON_SAMPLER_TYPE_MIN_P },
         { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_TEMPERATURE), COMMON_SAMPLER_TYPE_TEMPERATURE },
-        { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_XTC),         COMMON_SAMPLER_TYPE_XTC }
+        { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_XTC),         COMMON_SAMPLER_TYPE_XTC },
+        { common_sampler_type_to_chr(COMMON_SAMPLER_TYPE_INFILL),      COMMON_SAMPLER_TYPE_INFILL },
     };
 
     std::vector samplers;
diff --git a/examples/main/main.cpp b/examples/main/main.cpp
index fb10c20c5..65483c45f 100644
--- a/examples/main/main.cpp
+++ b/examples/main/main.cpp
@@ -569,30 +569,30 @@ int main(int argc, char ** argv) {
                     if (!params.ctx_shift){
                         LOG_DBG("\n\n%s: context full and context shift is disabled => stopping\n", __func__);
                         break;
-                    } else {
-                        if (params.n_predict == -2) {
-                            LOG_DBG("\n\n%s: context full and n_predict == -%d => stopping\n", __func__, params.n_predict);
-                            break;
-                        }
-
-                        const int n_left    = n_past - params.n_keep;
-                        const int n_discard = n_left/2;
-
-                        LOG_DBG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d, n_discard = %d\n",
-                                n_past, n_left, n_ctx, params.n_keep, n_discard);
-
-                        llama_kv_cache_seq_rm (ctx, 0, params.n_keep            , params.n_keep + n_discard);
-                        llama_kv_cache_seq_add(ctx, 0, params.n_keep + n_discard, n_past, -n_discard);
-
-                        n_past -= n_discard;
-
-                        LOG_DBG("after swap: n_past = %d\n", n_past);
-
-                        LOG_DBG("embd: %s\n", string_from(ctx, embd).c_str());
-
-                        LOG_DBG("clear session path\n");
-                        path_session.clear();
                     }
+
+                    if (params.n_predict == -2) {
+                        LOG_DBG("\n\n%s: context full and n_predict == -%d => stopping\n", __func__, params.n_predict);
+                        break;
+                    }
+
+                    const int n_left    = n_past - params.n_keep;
+                    const int n_discard = n_left/2;
+
+                    LOG_DBG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d, n_discard = %d\n",
+                            n_past, n_left, n_ctx, params.n_keep, n_discard);
+
+                    llama_kv_cache_seq_rm (ctx, 0, params.n_keep            , params.n_keep + n_discard);
+                    llama_kv_cache_seq_add(ctx, 0, params.n_keep + n_discard, n_past, -n_discard);
+
+                    n_past -= n_discard;
+
+                    LOG_DBG("after swap: n_past = %d\n", n_past);
+
+                    LOG_DBG("embd: %s\n", string_from(ctx, embd).c_str());
+
+                    LOG_DBG("clear session path\n");
+                    path_session.clear();
                 }
             } else {
                 // context extension via Self-Extend
diff --git a/include/llama.h b/include/llama.h
index 92d4c70c1..02bc7f087 100644
--- a/include/llama.h
+++ b/include/llama.h
@@ -953,6 +953,12 @@ extern "C" {
                                int32_t   lstrip,
                                   bool   special);
 
+    // check if token0 is contained as a prefix in token1
+    LLAMA_API bool llama_token_is_prefix(
+              const struct llama_model * model,
+                           llama_token   token0,
+                           llama_token   token1);
+
     /// @details Convert the provided tokens into text (inverse of llama_tokenize()).
     /// @param text The char pointer must be large enough to hold the resulting text.
     /// @return Returns the number of chars/bytes on success, no more than text_len_max.
@@ -1148,6 +1154,28 @@ extern "C" {
                              int32_t   n_logit_bias,
               const llama_logit_bias * logit_bias);
 
+    // this sampler is meant to be used for fill-in-the-middle infilling
+    // it's supposed to be used after top_k + top_p sampling
+    //
+    // 1. if the sum of the EOG probs times the number of candidates is higher than the sum of the other probs -> pick EOG
+    // 2. combine probs of tokens that have the same prefix
+    //
+    // example:
+    //
+    // - before:
+    //   "hel":   0.5
+    //   "hell":  0.2
+    //   "hello": 0.1
+    //   "dummy": 0.1
+    //
+    // - after:
+    //   "hel":   0.8
+    //   "dummy": 0.1
+    //
+    // 3. discard non-EOG tokens with low prob
+    // 4. if no tokens are left -> pick EOT
+    //
+    LLAMA_API struct llama_sampler * llama_sampler_init_infill(const struct llama_model * model);
 
     // Returns the seed used by the sampler if applicable, LLAMA_DEFAULT_SEED otherwise
     LLAMA_API uint32_t llama_sampler_get_seed(const struct llama_sampler * smpl);
diff --git a/src/llama-sampling.cpp b/src/llama-sampling.cpp
index 67a78c3ac..2e6550682 100644
--- a/src/llama-sampling.cpp
+++ b/src/llama-sampling.cpp
@@ -1739,6 +1739,207 @@ struct llama_sampler * llama_sampler_init_logit_bias(
     };
 }
 
+// infill
+
+//#define GGML_DEBUG_SAMPLER_INFILL
+
+struct llama_sampler_infill {
+    const struct llama_vocab * vocab;
+};
+
+static const char * llama_sampler_infill_name(const struct llama_sampler * /*smpl*/) {
+    return "infill";
+}
+
+static void llama_sampler_infill_apply(struct llama_sampler * smpl, llama_token_data_array * cur_p) {
+    auto * ctx = (llama_sampler_infill *) smpl->ctx;
+
+    llama_sampler_softmax_impl(cur_p);
+
+#if defined(GGML_DEBUG_SAMPLER_INFILL)
+#define LOG_DBG_CUR LLAMA_LOG_DEBUG
+#else
+#define LOG_DBG_CUR(...)
+#endif
+
+    for (size_t i = 0; i < cur_p->size; ++i) {
+        LOG_DBG_CUR("%s: cur_p[%3zu] = { id: %6d, p: %.6f, logit: %6.3f }\n", __func__, i, cur_p->data[i].id, cur_p->data[i].p, cur_p->data[i].logit);
+    }
+
+    float p_txt_sum = 0.0f;
+    float p_eog_sum = 0.0f;
+
+    for (size_t i = 0; i < cur_p->size; ++i) {
+        if (llama_token_is_eog_impl(*ctx->vocab, cur_p->data[i].id)) {
+            p_eog_sum += cur_p->data[i].p;
+        } else {
+            p_txt_sum += cur_p->data[i].p;
+        }
+    }
+
+    const float rat = p_eog_sum == 0.0 ? INFINITY : p_txt_sum / p_eog_sum; GGML_UNUSED(rat);
+
+    LOG_DBG_CUR("%s: p_txt_sum = %.2f, p_eog_sum = %.2f, rat = %.2f, n = %zu\n", __func__, p_txt_sum, p_eog_sum, rat, cur_p->size);
+
+    if (3*p_eog_sum*cur_p->size > p_txt_sum) {
+        LOG_DBG_CUR("%s: the ratio p_txt/p_eog = %.2f is too low -> sampling EOG\n", __func__, p_txt_sum/p_eog_sum);
+
+        // keep just the EOG tokens
+        const auto size_org = cur_p->size;
+
+        cur_p->size = 0;
+
+        float p_sum = 0.0f;
+
+        for (size_t i = 0; i < size_org; ++i) {
+            if (llama_token_is_eog_impl(*ctx->vocab, cur_p->data[i].id)) {
+                p_sum += cur_p->data[i].p;
+
+                cur_p->data[cur_p->size++] = cur_p->data[i];
+            }
+        }
+
+        // normalize probs
+        for (size_t i = 0; i < cur_p->size; ++i) {
+            cur_p->data[i].p /= p_sum;
+        }
+
+        return;
+    }
+
+    size_t n_combined = 0; GGML_UNUSED(n_combined);
+
+    // combine tokens with common prefix
+    for (size_t i = 0; i < cur_p->size; ++i) {
+        for (size_t j = 0; j < cur_p->size; ++j) {
+            if (cur_p->data[i].logit == -INFINITY) {
+                break;
+            }
+
+            if (i == j || cur_p->data[j].logit == -INFINITY) {
+                continue;
+            }
+
+            if (llama_token_is_prefix_impl(*ctx->vocab, cur_p->data[i].id, cur_p->data[j].id)) {
+                if (cur_p->data[i].p >  cur_p->data[j].p) {
+                    cur_p->data[i].p += cur_p->data[j].p;
+                    cur_p->data[j].logit = -INFINITY;
+                    cur_p->data[j].p     = 0.0f;
+                } else {
+                    cur_p->data[j].p += cur_p->data[i].p;
+                    cur_p->data[i].logit = -INFINITY;
+                    cur_p->data[i].p     = 0.0f;
+                }
+
+                n_combined++;
+            }
+        }
+    }
+
+    size_t n_non_eog = 0;
+
+    size_t size_org = cur_p->size;
+
+    float p_sum = 0.0f;
+    float thold = 0.2f;
+
+    cur_p->size = 0;
+
+    LOG_DBG_CUR("%s: n_combined = %zu, applying thold = %.3f\n", __func__, n_combined, thold);
+
+    for (size_t i = 0; i < size_org; ++i) {
+        const bool is_eog = llama_token_is_eog_impl(*ctx->vocab, cur_p->data[i].id);
+
+        if (cur_p->data[i].p < thold && !is_eog) {
+            continue;
+        }
+
+        if (!is_eog) {
+            ++n_non_eog;
+        }
+
+        p_sum += cur_p->data[i].p;
+
+        // keep this token
+        cur_p->data[cur_p->size++] = cur_p->data[i];
+    }
+
+    LOG_DBG_CUR("%s: n_non_eog = %zu\n", __func__, n_non_eog);
+
+    // if no non-EOG tokens are left -> reduce cur_p to single EOT token
+    if (n_non_eog == 0) {
+        cur_p->size = 1;
+        cur_p->data[0].id = llama_token_eot_impl(*ctx->vocab);
+        cur_p->data[0].logit = 1.0f;
+
+        return;
+    }
+
+    // normalize probs
+    for (size_t i = 0; i < cur_p->size; ++i) {
+        cur_p->data[i].p /= p_sum;
+
+        LOG_DBG_CUR("%s: cur_p[%3zu] = { id: %6d, p: %.6f, logit: %6.3f }\n", __func__, i, cur_p->data[i].id, cur_p->data[i].p, cur_p->data[i].logit);
+    }
+
+    size_org = cur_p->size;
+    p_sum = 0.0f;
+    thold = 1.0/(n_non_eog + 1);
+
+    cur_p->size = 0;
+
+    LOG_DBG_CUR("%s: applying thold = %.3f\n", __func__, thold);
+
+    for (size_t i = 0; i < size_org; ++i) {
+        const bool is_eog = llama_token_is_eog_impl(*ctx->vocab, cur_p->data[i].id);
+
+        if (cur_p->data[i].p < thold && !is_eog) {
+            continue;
+        }
+
+        p_sum += cur_p->data[i].p;
+
+        cur_p->data[cur_p->size++] = cur_p->data[i];
+    }
+
+    // normalize probs
+    for (size_t i = 0; i < cur_p->size; ++i) {
+        cur_p->data[i].p /= p_sum;
+
+        LOG_DBG_CUR("%s: cur_p[%3zu] = { id: %6d, p: %.6f, logit: %6.3f }\n", __func__, i, cur_p->data[i].id, cur_p->data[i].p, cur_p->data[i].logit);
+    }
+
+#undef LOG_DBG_CUR
+}
+
+static struct llama_sampler * llama_sampler_infill_clone(const struct llama_sampler * smpl) {
+    const auto * ctx = (const llama_sampler_infill *) smpl->ctx;
+    return llama_sampler_init_infill_impl(*ctx->vocab);
+}
+
+static void llama_sampler_infill_free(struct llama_sampler * smpl) {
+    delete (llama_sampler_infill *) smpl->ctx;
+}
+
+static struct llama_sampler_i llama_sampler_infill_i = {
+    /* .name   = */ llama_sampler_infill_name,
+    /* .accept = */ nullptr,
+    /* .apply  = */ llama_sampler_infill_apply,
+    /* .reset  = */ nullptr,
+    /* .clone  = */ llama_sampler_infill_clone,
+    /* .free   = */ llama_sampler_infill_free,
+};
+
+struct llama_sampler * llama_sampler_init_infill_impl(
+        const struct llama_vocab & vocab) {
+    return new llama_sampler {
+        /* .iface = */ &llama_sampler_infill_i,
+        /* .ctx   = */ new llama_sampler_infill {
+            /* .vocab = */ &vocab,
+        },
+    };
+}
+
 // utils
 
 uint32_t llama_sampler_get_seed(const struct llama_sampler * smpl) {
diff --git a/src/llama-sampling.h b/src/llama-sampling.h
index d90b14713..2683f1b92 100644
--- a/src/llama-sampling.h
+++ b/src/llama-sampling.h
@@ -4,8 +4,6 @@
 
 #include "llama-grammar.h"
 
-#include 
-
 struct llama_vocab;
 struct llama_grammar;
 
@@ -27,3 +25,6 @@ struct llama_sampler * llama_sampler_init_grammar_impl(
         const struct llama_vocab & vocab,
                       const char * grammar_str,
                       const char * grammar_root);
+
+struct llama_sampler * llama_sampler_init_infill_impl(
+        const struct llama_vocab & vocab);
diff --git a/src/llama-vocab.cpp b/src/llama-vocab.cpp
index a27394a37..070de9365 100644
--- a/src/llama-vocab.cpp
+++ b/src/llama-vocab.cpp
@@ -1858,6 +1858,23 @@ int32_t llama_token_to_piece_impl(const struct llama_vocab & vocab, llama_token
     return 0;
 }
 
+bool llama_token_is_prefix_impl(
+        const struct llama_vocab & vocab,
+                     llama_token   token0,
+                     llama_token   token1) {
+    char text_buf_0[128];
+    char text_buf_1[128];
+
+    const int32_t len0 = llama_token_to_piece_impl(vocab, token0, text_buf_0, sizeof(text_buf_0) - 1, 0, false);
+    const int32_t len1 = llama_token_to_piece_impl(vocab, token1, text_buf_1, sizeof(text_buf_1) - 1, 0, false);
+
+    if (len0 <= 0 || len1 <= 0) {
+        return false;
+    }
+
+    return len0 <= len1 && memcmp(text_buf_0, text_buf_1, len0) == 0;
+}
+
 int32_t llama_detokenize_impl(
         const struct llama_vocab & vocab,
                const llama_token * tokens,
diff --git a/src/llama-vocab.h b/src/llama-vocab.h
index 17e14488a..d958d0073 100644
--- a/src/llama-vocab.h
+++ b/src/llama-vocab.h
@@ -48,7 +48,7 @@ struct llama_vocab {
     id special_cls_id  = LLAMA_TOKEN_NULL;
     id special_mask_id = LLAMA_TOKEN_NULL;
 
-    id linefeed_id    = 13;
+    id linefeed_id = 13;
 
     // fim tokens
     id special_fim_pre_id = LLAMA_TOKEN_NULL;
@@ -149,6 +149,12 @@ int32_t llama_token_to_piece_impl(
                          int32_t   lstrip,
                             bool   special);
 
+// check if token0 is contained as a prefix in token1
+bool llama_token_is_prefix_impl(
+        const struct llama_vocab & vocab,
+                     llama_token   token0,
+                     llama_token   token1);
+
 int32_t llama_detokenize_impl(
         const struct llama_vocab & vocab,
                const llama_token * tokens,
diff --git a/src/llama.cpp b/src/llama.cpp
index 511f91802..8d44c73c8 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -21500,6 +21500,13 @@ int32_t llama_token_to_piece(
     return llama_token_to_piece_impl(model->vocab, token, buf, length, lstrip, special);
 }
 
+bool llama_token_is_prefix(
+    const struct llama_model * model,
+                 llama_token   token0,
+                 llama_token   token1) {
+    return llama_token_is_prefix_impl(model->vocab, token0, token1);
+}
+
 int32_t llama_detokenize(
     const struct llama_model * model,
            const llama_token * tokens,
@@ -21830,6 +21837,10 @@ struct llama_sampler * llama_sampler_init_grammar(const struct llama_model * mod
     return llama_sampler_init_grammar_impl(model->vocab, grammar_str, grammar_root);
 }
 
+struct llama_sampler * llama_sampler_init_infill(const struct llama_model * model) {
+    return llama_sampler_init_infill_impl(model->vocab);
+}
+
 //
 // model split
 //

From becfd387f6919d99ec34b76c2522f90ac250c489 Mon Sep 17 00:00:00 2001
From: leo-pony 
Date: Wed, 16 Oct 2024 08:51:46 +0800
Subject: [PATCH 17/36] [CANN] Fix cann compilation error (#9891)

Fix cann compilation error after merging llama.cpp supports dynamically loadable backends.
---
 ggml/src/ggml-cann.cpp | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/ggml/src/ggml-cann.cpp b/ggml/src/ggml-cann.cpp
index db5f8f186..ec3c0a688 100644
--- a/ggml/src/ggml-cann.cpp
+++ b/ggml/src/ggml-cann.cpp
@@ -1148,6 +1148,7 @@ ggml_backend_cann_buffer_type(int32_t device) {
         for (int32_t i = 0; i < GGML_CANN_MAX_DEVICES; i++) {
             ggml_backend_cann_buffer_types[i] = {
                 /* .iface    = */ ggml_backend_cann_buffer_type_interface,
+                /* .device    = */ nullptr,
                 /* .context  = */
                  new ggml_backend_cann_buffer_type_context{
                     i, "CANN" + std::to_string(i)},
@@ -1868,7 +1869,7 @@ static ggml_backend_event_t ggml_backend_cann_event_new(
     ACL_CHECK(aclrtCreateEvent(&event));
 
     return new ggml_backend_event{
-        /* .backend = */ backend,
+        /* .device = */ nullptr,
         /* .context = */ event,
     };
 }
@@ -1895,10 +1896,9 @@ static void ggml_backend_cann_event_free(ggml_backend_event_t event) {
  *
  * @param event Pointer to the event structure to be recorded.
  */
-static void ggml_backend_cann_event_record(ggml_backend_event_t event) {
+static void ggml_backend_cann_event_record(ggml_backend_t backend, ggml_backend_event_t event) {
     ggml_backend_cann_context* cann_ctx =
-        (ggml_backend_cann_context*)event->backend->context;
-
+        (ggml_backend_cann_context*)backend->context;
     ACL_CHECK(aclrtRecordEvent((aclrtEvent)event->context, cann_ctx->stream()));
 }
 
@@ -1916,8 +1916,7 @@ static void ggml_backend_cann_event_wait(ggml_backend_t backend,
                                          ggml_backend_event_t event) {
     ggml_backend_cann_context* cann_ctx =
         (ggml_backend_cann_context*)backend->context;
-
-    if (ggml_backend_is_cann(event->backend)) {
+    if (ggml_backend_is_cann(backend)) {
         ACL_CHECK(aclrtStreamWaitEvent(cann_ctx->stream(),
                                        (aclrtEvent)event->context));
     } else {

From cd60b88bf7ad7785fb6ac9864e360cf10e42faad Mon Sep 17 00:00:00 2001
From: Daniel Bevenius 
Date: Wed, 9 Oct 2024 16:40:35 +0200
Subject: [PATCH 18/36] ggml-alloc : remove buffer_id from leaf_alloc
 (ggml/987)

This commit removes the buffer_id field from the leaf_alloc struct.

The motivation for is that this field is only written to and never
read/used as far as I can tell. Each tensor_alloc has a buffer_id field
and this is what caused me to look into this more closely, to
understand what the buffer_id in leaf_alloc was used for.
---
 ggml/src/ggml-alloc.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/ggml/src/ggml-alloc.c b/ggml/src/ggml-alloc.c
index 28548fbbb..041de9e3e 100644
--- a/ggml/src/ggml-alloc.c
+++ b/ggml/src/ggml-alloc.c
@@ -348,7 +348,6 @@ struct tensor_alloc {
 };
 
 struct leaf_alloc {
-    int buffer_id;
     struct tensor_alloc leaf;
 };
 
@@ -740,7 +739,6 @@ bool ggml_gallocr_reserve_n(ggml_gallocr_t galloc, struct ggml_cgraph * graph, c
     for (int i = 0; i < graph->n_leafs; i++) {
         struct ggml_tensor * leaf = graph->leafs[i];
         struct hash_node * hn = ggml_gallocr_hash_get(galloc, leaf);
-        galloc->leaf_allocs[i].buffer_id = hn->buffer_id;
         if (leaf->view_src || leaf->data) {
             galloc->leaf_allocs[i].leaf.buffer_id = -1;
             galloc->leaf_allocs[i].leaf.offset = SIZE_MAX;

From 0e41b300ed28f7fe185d938b2e3d56a0bf7411ed Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Wed, 16 Oct 2024 11:28:14 +0300
Subject: [PATCH 19/36] sync : ggml

---
 scripts/sync-ggml.last | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/scripts/sync-ggml.last b/scripts/sync-ggml.last
index 3cca9cc2f..6d31b21b9 100644
--- a/scripts/sync-ggml.last
+++ b/scripts/sync-ggml.last
@@ -1 +1 @@
-564f42082f858f9674b2a2e06e9e779d9ed2c754
+2327bda7a55ac6b72614ac5ebd5c5a5e02553b9b

From 1f66b699c48cb5ab3265ed72c48e8549b1674291 Mon Sep 17 00:00:00 2001
From: Alexey Parfenov 
Date: Wed, 16 Oct 2024 08:35:53 +0000
Subject: [PATCH 20/36] server : fix the disappearance of the end of the text
 (#9867)

* server: fix the disappearance of the end of the text when streaming with stop strings

* simplify "send text" checks
---
 examples/server/server.cpp | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index d53cca84c..b5e63384c 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -1090,22 +1090,21 @@ struct server_context {
             size_t pos = std::min(slot.n_sent_text, slot.generated_text.size());
 
             const std::string str_test = slot.generated_text.substr(pos);
-            bool is_stop_full = false;
+            bool send_text = true;
 
             size_t stop_pos = slot.find_stopping_strings(str_test, token_str.size(), STOP_TYPE_FULL);
             if (stop_pos != std::string::npos) {
-                is_stop_full = true;
                 slot.generated_text.erase(
                     slot.generated_text.begin() + pos + stop_pos,
                     slot.generated_text.end());
                 pos = std::min(slot.n_sent_text, slot.generated_text.size());
-            } else {
-                is_stop_full = false;
+            } else if (slot.has_next_token) {
                 stop_pos = slot.find_stopping_strings(str_test, token_str.size(), STOP_TYPE_PARTIAL);
+                send_text = stop_pos == std::string::npos;
             }
 
             // check if there is any token to predict
-            if (stop_pos == std::string::npos || (!slot.has_next_token && !is_stop_full && stop_pos > 0)) {
+            if (send_text) {
                 // no send the stop word in the response
                 result.text_to_send = slot.generated_text.substr(pos, std::string::npos);
                 slot.n_sent_text += result.text_to_send.size();

From 10433e8b457c4cfd759cbb41fc55fc398db4a5da Mon Sep 17 00:00:00 2001
From: Molly Sophia 
Date: Wed, 16 Oct 2024 18:10:21 +0800
Subject: [PATCH 21/36] llama : add tensor name for "result_norm" (#9907)

Signed-off-by: Molly Sophia 
---
 src/llama.cpp | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/src/llama.cpp b/src/llama.cpp
index 8d44c73c8..c51b49c56 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -16095,9 +16095,11 @@ struct llm_build_context {
         cur = ggml_get_rows(ctx0, cur, inp_out_ids);
 
         cur = llm_build_norm(ctx0, cur, hparams, model.output_norm, model.output_norm_b, LLM_NORM, cb, -1);
-        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
+        cb(cur, "result_norm", -1);
 
+        cur = llm_build_lora_mm(lctx, ctx0, model.output, cur);
         cb(cur, "result_output", -1);
+
         ggml_build_forward_expand(gf, cur);
 
         return gf;

From 66c2c93082289325199ae1f773f3b0ab2e399a47 Mon Sep 17 00:00:00 2001
From: Joe Eli McIlvain 
Date: Wed, 16 Oct 2024 09:03:24 -0700
Subject: [PATCH 22/36] grammar : fix JSON Schema for string regex with
 top-level alt. (#9903)

Prior to this commit, using a JSON Schema containing a string
with `pattern` regular expression that uses top-level alternation
(e.g. `"pattern": "^A|B|C|D$"`) would result in invalid JSON
output from the constrained sampling grammar, because it
ended up creating a grammar rule like this for the string:

```
thing ::= "\"" "A" | "B" | "C" | "D" "\"" space
```

Note that this rule will only match a starting quote for the "A" case,
and will only match an ending quote for the "D" case,
so this rule will always produce invalid JSON when used for sampling
(that is, the JSON will always be lacking the starting quote,
the ending quote, or both).

This was fixed in a simple way by adding parentheses to the
generated rule (for all string pattern rules, to keep it simple),
such that the new generated rule looks like this (correct):

```
thing ::= "\"" ("A" | "B" | "C" | "D") "\"" space
```
---
 common/json-schema-to-grammar.cpp             |  2 +-
 examples/json_schema_to_grammar.py            |  2 +-
 .../server/public/json-schema-to-grammar.mjs  |  2 +-
 tests/test-json-schema-to-grammar.cpp         | 21 +++++++++++++++----
 4 files changed, 20 insertions(+), 7 deletions(-)

diff --git a/common/json-schema-to-grammar.cpp b/common/json-schema-to-grammar.cpp
index 881eb49e3..dadc18c8b 100644
--- a/common/json-schema-to-grammar.cpp
+++ b/common/json-schema-to-grammar.cpp
@@ -611,7 +611,7 @@ private:
             }
             return join_seq();
         };
-        return _add_rule(name, "\"\\\"\" " + to_rule(transform()) + " \"\\\"\" space");
+        return _add_rule(name, "\"\\\"\" (" + to_rule(transform()) + ") \"\\\"\" space");
     }
 
     /*
diff --git a/examples/json_schema_to_grammar.py b/examples/json_schema_to_grammar.py
index a8779bf3b..fc9f0097f 100755
--- a/examples/json_schema_to_grammar.py
+++ b/examples/json_schema_to_grammar.py
@@ -540,7 +540,7 @@ class SchemaConverter:
         return self._add_rule(
             name,
             to_rule(transform()) if self._raw_pattern \
-                else "\"\\\"\" " + to_rule(transform()) + " \"\\\"\" space")
+                else "\"\\\"\" (" + to_rule(transform()) + ") \"\\\"\" space")
 
 
     def _resolve_ref(self, ref):
diff --git a/examples/server/public/json-schema-to-grammar.mjs b/examples/server/public/json-schema-to-grammar.mjs
index 7267f3f9c..e67bb15c1 100644
--- a/examples/server/public/json-schema-to-grammar.mjs
+++ b/examples/server/public/json-schema-to-grammar.mjs
@@ -529,7 +529,7 @@ export class SchemaConverter {
       return joinSeq();
     };
 
-    return this._addRule(name, "\"\\\"\" " + toRule(transform()) + " \"\\\"\" space")
+    return this._addRule(name, "\"\\\"\" (" + toRule(transform()) + ") \"\\\"\" space")
   }
 
   _notStrings(strings) {
diff --git a/tests/test-json-schema-to-grammar.cpp b/tests/test-json-schema-to-grammar.cpp
index 3a89598a8..9d2db91f5 100755
--- a/tests/test-json-schema-to-grammar.cpp
+++ b/tests/test-json-schema-to-grammar.cpp
@@ -696,7 +696,7 @@ static void test_all(const std::string & lang, std::function
Date: Wed, 16 Oct 2024 19:24:05 +0200
Subject: [PATCH 23/36] llava : fix typo in error message [no ci] (#9884)

---
 examples/llava/llava.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/examples/llava/llava.cpp b/examples/llava/llava.cpp
index 8558c6bdc..2c96973c8 100644
--- a/examples/llava/llava.cpp
+++ b/examples/llava/llava.cpp
@@ -432,7 +432,7 @@ struct llava_image_embed * llava_image_embed_make_with_bytes(struct clip_ctx * c
     bool image_embed_result = llava_image_embed_make_with_clip_img(ctx_clip, n_threads, img, &image_embed, &n_image_pos);
     if (!image_embed_result) {
         clip_image_u8_free(img);
-        LOG_ERR("%s: coulnd't embed the image\n", __func__);
+        LOG_ERR("%s: couldn't embed the image\n", __func__);
         return NULL;
     }
 

From 9e041024481f6b249ab8918e18b9477f873b5a5e Mon Sep 17 00:00:00 2001
From: Daniel Bevenius 
Date: Wed, 16 Oct 2024 19:34:28 +0200
Subject: [PATCH 24/36] llama : suppress conversion from 'size_t' to 'int'
 (#9046)

* llama : suppress conversion from 'size_t' to 'int'

This commit updates llm_tokenizer_spm.tokenize to suppress/remove the
following warnings that are generated on Windows when using MSVC:

```console
src\llama-vocab.cpp(211,1): warning C4267: 'argument':
    conversion from 'size_t' to 'int', possible loss of data
src\llama-vocab.cpp(517,1): warning C4267: 'argument':
    conversion from 'size_t' to 'int', possible loss of data
```

This is done by adding a cast for the size_t returned from
symbols.size(). I believe this is safe as it seems unlikely that
symbols, which stores an entry for each UTF8 character, would become
larger than INT_MAX.

The motivation for this change is to reduce the number of warnings that
are currently generated when building on Windows.

* squash! llama : suppress conversion from 'size_t' to 'int'

Move cast into for loop.
---
 src/llama-vocab.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/llama-vocab.cpp b/src/llama-vocab.cpp
index 070de9365..57d56a3d3 100644
--- a/src/llama-vocab.cpp
+++ b/src/llama-vocab.cpp
@@ -221,7 +221,7 @@ struct llm_tokenizer_spm_session {
         }
 
         // seed the work queue with all possible 2-character tokens.
-        for (size_t i = 1; i < symbols.size(); ++i) {
+        for (int i = 1; i < (int) symbols.size(); ++i) {
             try_add_bigram(i - 1, i);
         }
 
@@ -563,7 +563,7 @@ struct llm_tokenizer_bpe_session {
                 index++;
                 symbols.emplace_back(sym);
             }
-            for (size_t i = 1; i < symbols.size(); ++i) {
+            for (int i = 1; i < (int) symbols.size(); ++i) {
                 add_new_bigram(i - 1, i);
             }
 

From 73afe681aa76e818733fc1f30de082c1d6910bcd Mon Sep 17 00:00:00 2001
From: "Gilad S." <7817232+giladgd@users.noreply.github.com>
Date: Thu, 17 Oct 2024 01:36:51 +0300
Subject: [PATCH 25/36] fix: use `vm_allocate` to allocate CPU backend buffer
 on macOS (#9875)

* fix: use `vm_allocate` to allocate CPU backend buffer on macOS

* fix: switch to `posix_memalign` to keep existing `free()` usages work

* feat: move `GGML_ALIGNED_MALLOC` to `ggml-backend-impl.h`, add support for `vm_allocate` on macOS

* style: formatting

* fix: move const outside of `#ifndef`

* style: formatting

* fix: unused var

* fix: transform `GGML_ALIGNED_MALLOC` and `GGML_ALIGNED_FREE` into functions and add them to `ggml-impl.h`

* fix: unused var

* fix: page align to `GGUF_DEFAULT_ALIGNMENT`

* fix: page align to `TENSOR_ALIGNMENT`

* fix: convert `TENSOR_ALIGNMENT` to a macro

* fix: increase page size to `32` on iOS

* fix: iOS page size

* fix: `hbw_posix_memalign` alignment
---
 ggml/src/ggml-backend.cpp |  8 ++---
 ggml/src/ggml-impl.h      |  8 +++++
 ggml/src/ggml.c           | 74 +++++++++++++++++++++++++++------------
 3 files changed, 63 insertions(+), 27 deletions(-)

diff --git a/ggml/src/ggml-backend.cpp b/ggml/src/ggml-backend.cpp
index 15d650150..6d6ffeb4e 100644
--- a/ggml/src/ggml-backend.cpp
+++ b/ggml/src/ggml-backend.cpp
@@ -682,8 +682,6 @@ ggml_backend_t ggml_backend_init_best(void) {
 
 // backend CPU
 
-static const size_t TENSOR_ALIGNMENT = 32; // required for mmap as gguf only guarantees 32-byte alignment
-
 static const char * ggml_backend_cpu_buffer_get_name(ggml_backend_buffer_t buffer) {
     return "CPU";
 
@@ -702,7 +700,7 @@ static void * ggml_backend_cpu_buffer_get_base(ggml_backend_buffer_t buffer) {
 }
 
 static void ggml_backend_cpu_buffer_free_buffer(ggml_backend_buffer_t buffer) {
-    free(buffer->context);
+    ggml_aligned_free(buffer->context, buffer->size);
 }
 
 static void ggml_backend_cpu_buffer_memset_tensor(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, uint8_t value, size_t offset, size_t size) {
@@ -770,8 +768,8 @@ static const char * ggml_backend_cpu_buffer_type_get_name(ggml_backend_buffer_ty
 }
 
 static ggml_backend_buffer_t ggml_backend_cpu_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
-    size += TENSOR_ALIGNMENT;   // malloc may return an address that is not aligned
-    void * data = malloc(size); // TODO: use GGML_ALIGNED_MALLOC (move to ggml-impl.h)
+    void * data = ggml_aligned_malloc(size);
+
     if (data == NULL) {
         GGML_LOG_ERROR("%s: failed to allocate buffer of size %zu\n", __func__, size);
         return NULL;
diff --git a/ggml/src/ggml-impl.h b/ggml/src/ggml-impl.h
index d3f4bad8c..65c4f8119 100644
--- a/ggml/src/ggml-impl.h
+++ b/ggml/src/ggml-impl.h
@@ -19,6 +19,9 @@ extern "C" {
 #define MIN(a, b) ((a) < (b) ? (a) : (b))
 #define MAX(a, b) ((a) > (b) ? (a) : (b))
 
+// required for mmap as gguf only guarantees 32-byte alignment
+#define TENSOR_ALIGNMENT 32
+
 // static_assert should be a #define, but if it's not,
 // fall back to the _Static_assert C11 keyword.
 // if C99 - static_assert is noop
@@ -196,6 +199,11 @@ struct ggml_cgraph {
 
 struct ggml_cgraph ggml_graph_view(struct ggml_cgraph * cgraph, int i0, int i1);
 
+// Memory allocation
+
+void * ggml_aligned_malloc(size_t size);
+void ggml_aligned_free(void * ptr, size_t size);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index 3f01092d9..779b38d12 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -35,10 +35,6 @@
 #include 
 #endif
 
-#ifdef GGML_USE_METAL
-#include 
-#endif
-
 #if defined(__ARM_FEATURE_SVE) || defined(__ARM_FEATURE_MATMUL_INT8)
 #undef GGML_USE_LLAMAFILE
 #endif
@@ -189,6 +185,8 @@ typedef pthread_t ggml_thread_t;
 #endif
 
 #if defined(__APPLE__)
+#include 
+#include 
 #include 
 #endif
 
@@ -386,22 +384,40 @@ void ggml_log_callback_default(enum ggml_log_level level, const char * text, voi
 //#define GGML_SOFT_MAX_ACCELERATE
 #endif
 
+
+void * ggml_aligned_malloc(size_t size) {
 #if defined(_MSC_VER) || defined(__MINGW32__)
-#define GGML_ALIGNED_MALLOC(size) _aligned_malloc(size, GGML_MEM_ALIGN)
-#define GGML_ALIGNED_FREE(ptr)    _aligned_free(ptr)
+    return _aligned_malloc(size, TENSOR_ALIGNMENT);
 #else
-inline static void * ggml_aligned_malloc(size_t size) {
     if (size == 0) {
         GGML_LOG_WARN("Behavior may be unexpected when allocating 0 bytes for ggml_aligned_malloc!\n");
         return NULL;
     }
     void * aligned_memory = NULL;
 #ifdef GGML_USE_CPU_HBM
-    int result = hbw_posix_memalign(&aligned_memory, 16, size);
+    int result = hbw_posix_memalign(&aligned_memory, TENSOR_ALIGNMENT, size);
+#elif TARGET_OS_OSX
+    kern_return_t alloc_status = vm_allocate((vm_map_t) mach_task_self(), (vm_address_t *) &aligned_memory, size, VM_FLAGS_ANYWHERE);
+    int result = EFAULT;
+    switch (alloc_status) {
+        case KERN_SUCCESS:
+            result = 0;
+            break;
+        case KERN_INVALID_ADDRESS:
+            result = EINVAL;
+            break;
+        case KERN_NO_SPACE:
+            result = ENOMEM;
+            break;
+        default:
+            result = EFAULT;
+            break;
+    }
 #elif GGML_USE_METAL
-    int result = posix_memalign(&aligned_memory, sysconf(_SC_PAGESIZE), size);
+    const long page_size = sysconf(_SC_PAGESIZE);
+    int result = posix_memalign(&aligned_memory, MAX(TENSOR_ALIGNMENT, page_size), size);
 #else
-    int result = posix_memalign(&aligned_memory, GGML_MEM_ALIGN, size);
+    int result = posix_memalign(&aligned_memory, TENSOR_ALIGNMENT, size);
 #endif
     if (result != 0) {
         // Handle allocation failure
@@ -419,14 +435,26 @@ inline static void * ggml_aligned_malloc(size_t size) {
         return NULL;
     }
     return aligned_memory;
+#endif
 }
-#define GGML_ALIGNED_MALLOC(size) ggml_aligned_malloc(size)
-#ifdef GGML_USE_CPU_HBM
-#define GGML_ALIGNED_FREE(ptr)    if(NULL != ptr) hbw_free(ptr)
+
+void ggml_aligned_free(void * ptr, size_t size) {
+    GGML_UNUSED(size);
+#if defined(_MSC_VER) || defined(__MINGW32__)
+    _aligned_free(ptr);
+#elif GGML_USE_CPU_HBM
+    if (ptr != NULL) {
+        hbw_free(ptr);
+    }
+#elif TARGET_OS_OSX
+    if (ptr != NULL) {
+        vm_deallocate((vm_map_t)mach_task_self(), (vm_address_t)ptr, size);
+    }
 #else
-#define GGML_ALIGNED_FREE(ptr)    free(ptr)
-#endif
+    free(ptr);
 #endif
+}
+
 
 inline static void * ggml_malloc(size_t size) {
     if (size == 0) {
@@ -3869,7 +3897,7 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
 
     *ctx = (struct ggml_context) {
         /*.mem_size           =*/ mem_size,
-        /*.mem_buffer         =*/ params.mem_buffer ? params.mem_buffer : GGML_ALIGNED_MALLOC(mem_size),
+        /*.mem_buffer         =*/ params.mem_buffer ? params.mem_buffer : ggml_aligned_malloc(mem_size),
         /*.mem_buffer_owned   =*/ params.mem_buffer ? false : true,
         /*.no_alloc           =*/ params.no_alloc,
         /*.no_alloc_save      =*/ params.no_alloc,
@@ -3909,7 +3937,7 @@ void ggml_free(struct ggml_context * ctx) {
                     __func__, i, ggml_used_mem(ctx));
 
             if (ctx->mem_buffer_owned) {
-                GGML_ALIGNED_FREE(ctx->mem_buffer);
+                ggml_aligned_free(ctx->mem_buffer, ctx->mem_size);
             }
 
             found = true;
@@ -19608,9 +19636,10 @@ static void ggml_thread_cpumask_next(const bool * global_mask, bool * local_mask
 void ggml_threadpool_free(struct ggml_threadpool* threadpool) {
     if (!threadpool) return;
 
+    const int n_threads = threadpool->n_threads_max;
+
 #ifndef GGML_USE_OPENMP
     struct ggml_compute_state* workers = threadpool->workers;
-    const int n_threads = threadpool->n_threads_max;
 
     ggml_mutex_lock(&threadpool->mutex);
 
@@ -19630,8 +19659,9 @@ void ggml_threadpool_free(struct ggml_threadpool* threadpool) {
     ggml_cond_destroy(&threadpool->cond);
 #endif // GGML_USE_OPENMP
 
-    GGML_ALIGNED_FREE(threadpool->workers);
-    GGML_ALIGNED_FREE(threadpool);
+    const size_t workers_size = sizeof(struct ggml_compute_state) * n_threads;
+    ggml_aligned_free(threadpool->workers, workers_size);
+    ggml_aligned_free(threadpool, sizeof(struct ggml_threadpool));
 }
 
 #ifndef GGML_USE_OPENMP
@@ -20063,7 +20093,7 @@ static struct ggml_threadpool * ggml_threadpool_new_impl(
                 struct ggml_cplan * cplan) {
 
     struct ggml_threadpool * threadpool =
-        GGML_ALIGNED_MALLOC(sizeof(struct ggml_threadpool));
+        ggml_aligned_malloc(sizeof(struct ggml_threadpool));
     {
         threadpool->cgraph           = cgraph;
         threadpool->cplan            = cplan;
@@ -20084,7 +20114,7 @@ static struct ggml_threadpool * ggml_threadpool_new_impl(
 
     // Allocate and init workers state
     const size_t workers_size = sizeof(struct ggml_compute_state) * tpp->n_threads;
-    struct ggml_compute_state * workers = GGML_ALIGNED_MALLOC(workers_size);
+    struct ggml_compute_state * workers = ggml_aligned_malloc(workers_size);
 
     memset(workers, 0, workers_size);
     for (int j = 0; j < tpp->n_threads; j++) {

From 21942002780352b4a54f4bd3e5eefa3bc7f14fe6 Mon Sep 17 00:00:00 2001
From: "Gilad S." <7817232+giladgd@users.noreply.github.com>
Date: Thu, 17 Oct 2024 02:34:22 +0300
Subject: [PATCH 26/36] fix: allocating CPU buffer with size `0` (#9917)

---
 ggml/src/ggml-backend.cpp | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/ggml/src/ggml-backend.cpp b/ggml/src/ggml-backend.cpp
index 6d6ffeb4e..4b9bac21d 100644
--- a/ggml/src/ggml-backend.cpp
+++ b/ggml/src/ggml-backend.cpp
@@ -768,14 +768,19 @@ static const char * ggml_backend_cpu_buffer_type_get_name(ggml_backend_buffer_ty
 }
 
 static ggml_backend_buffer_t ggml_backend_cpu_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
-    void * data = ggml_aligned_malloc(size);
+    auto alloc_size = size;
+    if (alloc_size == 0) {
+        alloc_size = 1;
+    }
+
+    void * data = ggml_aligned_malloc(alloc_size);
 
     if (data == NULL) {
-        GGML_LOG_ERROR("%s: failed to allocate buffer of size %zu\n", __func__, size);
+        GGML_LOG_ERROR("%s: failed to allocate buffer of size %zu\n", __func__, alloc_size);
         return NULL;
     }
 
-    return ggml_backend_buffer_init(buft, ggml_backend_cpu_buffer_i, data, size);
+    return ggml_backend_buffer_init(buft, ggml_backend_cpu_buffer_i, data, alloc_size);
 }
 
 static size_t ggml_backend_cpu_buffer_type_get_alignment(ggml_backend_buffer_type_t buft) {

From f010b77a372ffcfaf4338c670d6d3ecd89aa4eb6 Mon Sep 17 00:00:00 2001
From: Diego Devesa 
Date: Thu, 17 Oct 2024 02:46:58 +0200
Subject: [PATCH 27/36] vulkan : add backend registry / device interfaces
 (#9721)

* vulkan : add backend registry / device interfaces

* llama : print devices used on model load
---
 ggml/include/ggml-vulkan.h |   2 +
 ggml/src/ggml-backend.cpp  |   9 +-
 ggml/src/ggml-vulkan.cpp   | 284 +++++++++++++++++++++++++++----------
 src/llama.cpp              |  54 ++-----
 4 files changed, 226 insertions(+), 123 deletions(-)

diff --git a/ggml/include/ggml-vulkan.h b/ggml/include/ggml-vulkan.h
index e074042ef..c03bbfe5e 100644
--- a/ggml/include/ggml-vulkan.h
+++ b/ggml/include/ggml-vulkan.h
@@ -24,6 +24,8 @@ GGML_API ggml_backend_buffer_type_t ggml_backend_vk_buffer_type(size_t dev_num);
 // pinned host buffer for use with the CPU backend for faster copies between CPU and GPU
 GGML_API ggml_backend_buffer_type_t ggml_backend_vk_host_buffer_type(void);
 
+GGML_API ggml_backend_reg_t ggml_backend_vk_reg(void);
+
 #ifdef  __cplusplus
 }
 #endif
diff --git a/ggml/src/ggml-backend.cpp b/ggml/src/ggml-backend.cpp
index 4b9bac21d..a3bc79a46 100644
--- a/ggml/src/ggml-backend.cpp
+++ b/ggml/src/ggml-backend.cpp
@@ -538,6 +538,10 @@ void * ggml_backend_reg_get_proc_address(ggml_backend_reg_t reg, const char * na
 #include "ggml-metal.h"
 #endif
 
+#ifdef GGML_USE_VULKAN
+#include "ggml-vulkan.h"
+#endif
+
 #ifdef GGML_USE_BLAS
 #include "ggml-blas.h"
 #endif
@@ -557,6 +561,9 @@ struct ggml_backend_registry {
 #ifdef GGML_USE_METAL
         register_backend(ggml_backend_metal_reg());
 #endif
+#ifdef GGML_USE_VULKAN
+        register_backend(ggml_backend_vk_reg());
+#endif
 #ifdef GGML_USE_BLAS
         register_backend(ggml_backend_blas_reg());
 #endif
@@ -564,7 +571,7 @@ struct ggml_backend_registry {
         register_backend(ggml_backend_rpc_reg());
 #endif
 
-        // TODO: sycl, vulkan, kompute, cann
+        // TODO: sycl, kompute, cann
 
         register_backend(ggml_backend_cpu_reg());
     }
diff --git a/ggml/src/ggml-vulkan.cpp b/ggml/src/ggml-vulkan.cpp
index 374c6ecd7..e749bbe70 100644
--- a/ggml/src/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan.cpp
@@ -1941,7 +1941,7 @@ static vk_device ggml_vk_get_device(size_t idx) {
         if (device->fp16) {
             device_extensions.push_back("VK_KHR_shader_float16_int8");
         }
-        device->name = device->properties.deviceName.data();
+        device->name = GGML_VK_NAME + std::to_string(idx);
 
         device_create_info = {
             vk::DeviceCreateFlags(),
@@ -1968,7 +1968,7 @@ static vk_device ggml_vk_get_device(size_t idx) {
 
         device->buffer_type = {
             /* .iface    = */ ggml_backend_vk_buffer_type_interface,
-            /* .device   = */ nullptr,
+            /* .device   = */ ggml_backend_reg_dev_get(ggml_backend_vk_reg(), idx),
             /* .context  = */ new ggml_backend_vk_buffer_type_context{ device->name, device },
         };
 
@@ -6378,7 +6378,7 @@ ggml_backend_buffer_type_t ggml_backend_vk_host_buffer_type() {
             /* .get_alloc_size   = */ ggml_backend_cpu_buffer_type()->iface.get_alloc_size,
             /* .is_host          = */ ggml_backend_cpu_buffer_type()->iface.is_host,
         },
-        /* .device   = */ nullptr,
+        /* .device   = */ ggml_backend_reg_dev_get(ggml_backend_vk_reg(), 0),
         /* .context  = */ nullptr,
     };
 
@@ -6581,9 +6581,135 @@ static ggml_status ggml_backend_vk_graph_compute(ggml_backend_t backend, ggml_cg
     UNUSED(backend);
 }
 
-static bool ggml_backend_vk_supports_op(ggml_backend_t backend, const ggml_tensor * op) {
-    // ggml_backend_vk_context * ctx = (ggml_backend_vk_context *) backend->context;
+// TODO: enable async and synchronize
+static ggml_backend_i ggml_backend_vk_interface = {
+    /* .get_name                = */ ggml_backend_vk_name,
+    /* .free                    = */ ggml_backend_vk_free,
+    /* .get_default_buffer_type = */ ggml_backend_vk_get_default_buffer_type,
+    /* .set_tensor_async        = */ NULL,  // ggml_backend_vk_set_tensor_async,
+    /* .get_tensor_async        = */ NULL,  // ggml_backend_vk_get_tensor_async,
+    /* .cpy_tensor_async        = */ NULL,  // ggml_backend_vk_cpy_tensor_async,
+    /* .synchronize             = */ NULL,  // ggml_backend_vk_synchronize,
+    /* .graph_plan_create       = */ NULL,
+    /* .graph_plan_free         = */ NULL,
+    /* .graph_plan_update       = */ NULL,
+    /* .graph_plan_compute      = */ NULL,
+    /* .graph_compute           = */ ggml_backend_vk_graph_compute,
+    /* .supports_op             = */ NULL,
+    /* .supports_buft           = */ NULL,
+    /* .offload_op              = */ NULL,
+    /* .event_record            = */ NULL,
+    /* .event_wait              = */ NULL,
+};
 
+static ggml_guid_t ggml_backend_vk_guid() {
+    static ggml_guid guid = { 0xb8, 0xf7, 0x4f, 0x86, 0x40, 0x3c, 0xe1, 0x02, 0x91, 0xc8, 0xdd, 0xe9, 0x02, 0x3f, 0xc0, 0x2b };
+    return &guid;
+}
+
+ggml_backend_t ggml_backend_vk_init(size_t dev_num) {
+    VK_LOG_DEBUG("ggml_backend_vk_init(" << dev_num << ")");
+
+    ggml_backend_vk_context * ctx = new ggml_backend_vk_context;
+    ggml_vk_init(ctx, dev_num);
+
+    ggml_backend_t vk_backend = new ggml_backend {
+        /* .guid      = */ ggml_backend_vk_guid(),
+        /* .interface = */ ggml_backend_vk_interface,
+        /* .device    = */ ggml_backend_reg_dev_get(ggml_backend_vk_reg(), dev_num),
+        /* .context   = */ ctx,
+    };
+
+    return vk_backend;
+}
+
+bool ggml_backend_is_vk(ggml_backend_t backend) {
+    return backend != NULL && ggml_guid_matches(backend->guid, ggml_backend_vk_guid());
+}
+
+int ggml_backend_vk_get_device_count() {
+    return ggml_vk_get_device_count();
+}
+
+void ggml_backend_vk_get_device_description(int device, char * description, size_t description_size) {
+    GGML_ASSERT(device < (int) vk_instance.device_indices.size());
+    int dev_idx = vk_instance.device_indices[device];
+    ggml_vk_get_device_description(dev_idx, description, description_size);
+}
+
+void ggml_backend_vk_get_device_memory(int device, size_t * free, size_t * total) {
+    GGML_ASSERT(device < (int) vk_instance.device_indices.size());
+
+    vk::PhysicalDevice vkdev = vk_instance.instance.enumeratePhysicalDevices()[vk_instance.device_indices[device]];
+
+    vk::PhysicalDeviceMemoryProperties memprops = vkdev.getMemoryProperties();
+
+    for (const vk::MemoryHeap& heap : memprops.memoryHeaps) {
+        if (heap.flags & vk::MemoryHeapFlagBits::eDeviceLocal) {
+            *total = heap.size;
+            *free = heap.size;
+            break;
+        }
+    }
+}
+
+//////////////////////////
+
+struct ggml_backend_vk_device_context {
+    int device;
+    std::string name;
+    std::string description;
+};
+
+static const char * ggml_backend_vk_device_get_name(ggml_backend_dev_t dev) {
+    ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)dev->context;
+    return ctx->name.c_str();
+}
+
+static const char * ggml_backend_vk_device_get_description(ggml_backend_dev_t dev) {
+    ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)dev->context;
+    return ctx->description.c_str();
+}
+
+static void ggml_backend_vk_device_get_memory(ggml_backend_dev_t device, size_t * free, size_t * total) {
+    ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)device->context;
+    ggml_backend_vk_get_device_memory(ctx->device, free, total);
+}
+
+static ggml_backend_buffer_type_t ggml_backend_vk_device_get_buffer_type(ggml_backend_dev_t dev) {
+    ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)dev->context;
+    return ggml_backend_vk_buffer_type(ctx->device);
+}
+
+static ggml_backend_buffer_type_t ggml_backend_vk_device_get_host_buffer_type(ggml_backend_dev_t dev) {
+    UNUSED(dev);
+    return ggml_backend_vk_host_buffer_type();
+}
+
+static enum ggml_backend_dev_type ggml_backend_vk_device_get_type(ggml_backend_dev_t dev) {
+    UNUSED(dev);
+    return GGML_BACKEND_DEVICE_TYPE_GPU_FULL;
+}
+
+static void ggml_backend_vk_device_get_props(ggml_backend_dev_t dev, struct ggml_backend_dev_props * props) {
+    props->name        = ggml_backend_vk_device_get_name(dev);
+    props->description = ggml_backend_vk_device_get_description(dev);
+    props->type        = ggml_backend_vk_device_get_type(dev);
+    ggml_backend_vk_device_get_memory(dev, &props->memory_free, &props->memory_total);
+    props->caps = {
+        /* async       */ false,
+        /* host_buffer */ true,
+        /* events      */ false,
+    };
+}
+
+static ggml_backend_t ggml_backend_vk_device_init(ggml_backend_dev_t dev, const char * params) {
+    UNUSED(params);
+    ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)dev->context;
+    return ggml_backend_vk_init(ctx->device);
+}
+
+static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggml_tensor * op) {
     switch (op->op) {
         case GGML_OP_UNARY:
             switch (ggml_get_unary_op(op)) {
@@ -6701,97 +6827,101 @@ static bool ggml_backend_vk_supports_op(ggml_backend_t backend, const ggml_tenso
             return false;
     }
 
-    UNUSED(backend);
+    UNUSED(dev);
 }
 
-static bool ggml_backend_vk_offload_op(ggml_backend_t backend, const ggml_tensor * op) {
+static bool ggml_backend_vk_device_supports_buft(ggml_backend_dev_t dev, ggml_backend_buffer_type_t buft) {
+    if (buft->iface.get_name != ggml_backend_vk_buffer_type_name) {
+        return false;
+    }
+
+    ggml_backend_vk_device_context * ctx = (ggml_backend_vk_device_context *)dev->context;
+    ggml_backend_vk_buffer_type_context * buft_ctx = (ggml_backend_vk_buffer_type_context *)buft->context;
+
+    return buft_ctx->device->idx == ctx->device;
+}
+
+static bool ggml_backend_vk_device_offload_op(ggml_backend_dev_t dev, const ggml_tensor * op) {
     const int min_batch_size = 32;
 
     return (op->ne[1] >= min_batch_size && op->op != GGML_OP_GET_ROWS) ||
            (op->ne[2] >= min_batch_size && op->op == GGML_OP_MUL_MAT_ID);
 
-    UNUSED(backend);
+    UNUSED(dev);
 }
 
-static bool ggml_backend_vk_supports_buft(ggml_backend_t backend, ggml_backend_buffer_type_t buft) {
-    if (buft->iface.get_name != ggml_backend_vk_buffer_type_name) {
-        return false;
-    }
-
-    ggml_backend_vk_buffer_type_context * buft_ctx = (ggml_backend_vk_buffer_type_context *)buft->context;
-    ggml_backend_vk_context * ctx = (ggml_backend_vk_context *)backend->context;
-
-    return buft_ctx->device == ctx->device;
-}
-
-// TODO: enable async and synchronize
-static ggml_backend_i ggml_backend_vk_interface = {
-    /* .get_name                = */ ggml_backend_vk_name,
-    /* .free                    = */ ggml_backend_vk_free,
-    /* .get_default_buffer_type = */ ggml_backend_vk_get_default_buffer_type,
-    /* .set_tensor_async        = */ NULL,  // ggml_backend_vk_set_tensor_async,
-    /* .get_tensor_async        = */ NULL,  // ggml_backend_vk_get_tensor_async,
-    /* .cpy_tensor_async        = */ NULL,  // ggml_backend_vk_cpy_tensor_async,
-    /* .synchronize             = */ NULL,  // ggml_backend_vk_synchronize,
-    /* .graph_plan_create       = */ NULL,
-    /* .graph_plan_free         = */ NULL,
-    /* .graph_plan_update       = */ NULL,
-    /* .graph_plan_compute      = */ NULL,
-    /* .graph_compute           = */ ggml_backend_vk_graph_compute,
-    /* .supports_op             = */ ggml_backend_vk_supports_op,
-    /* .supports_buft           = */ ggml_backend_vk_supports_buft,
-    /* .offload_op              = */ ggml_backend_vk_offload_op,
-    /* .event_record            = */ NULL,
-    /* .event_wait              = */ NULL,
+static const struct ggml_backend_device_i ggml_backend_vk_device_i = {
+    /* .get_name             = */ ggml_backend_vk_device_get_name,
+    /* .get_description      = */ ggml_backend_vk_device_get_description,
+    /* .get_memory           = */ ggml_backend_vk_device_get_memory,
+    /* .get_type             = */ ggml_backend_vk_device_get_type,
+    /* .get_props            = */ ggml_backend_vk_device_get_props,
+    /* .init_backend         = */ ggml_backend_vk_device_init,
+    /* .get_buffer_type      = */ ggml_backend_vk_device_get_buffer_type,
+    /* .get_host_buffer_type = */ ggml_backend_vk_device_get_host_buffer_type,
+    /* .buffer_from_host_ptr = */ NULL,
+    /* .supports_op          = */ ggml_backend_vk_device_supports_op,
+    /* .supports_buft        = */ ggml_backend_vk_device_supports_buft,
+    /* .offload_op           = */ ggml_backend_vk_device_offload_op,
+    /* .event_new            = */ NULL,
+    /* .event_free           = */ NULL,
+    /* .event_synchronize    = */ NULL,
 };
 
-static ggml_guid_t ggml_backend_vk_guid() {
-    static ggml_guid guid = { 0xb8, 0xf7, 0x4f, 0x86, 0x40, 0x3c, 0xe1, 0x02, 0x91, 0xc8, 0xdd, 0xe9, 0x02, 0x3f, 0xc0, 0x2b };
-    return &guid;
+static const char * ggml_backend_vk_reg_get_name(ggml_backend_reg_t reg) {
+    UNUSED(reg);
+    return GGML_VK_NAME;
 }
 
-ggml_backend_t ggml_backend_vk_init(size_t dev_num) {
-    VK_LOG_DEBUG("ggml_backend_vk_init(" << dev_num << ")");
-
-    ggml_backend_vk_context * ctx = new ggml_backend_vk_context;
-    ggml_vk_init(ctx, dev_num);
-
-    ggml_backend_t vk_backend = new ggml_backend {
-        /* .guid      = */ ggml_backend_vk_guid(),
-        /* .interface = */ ggml_backend_vk_interface,
-        /* .device    = */ nullptr,
-        /* .context   = */ ctx,
-    };
-
-    return vk_backend;
+static size_t ggml_backend_vk_reg_get_device_count(ggml_backend_reg_t reg) {
+    UNUSED(reg);
+    return ggml_backend_vk_get_device_count();
 }
 
-bool ggml_backend_is_vk(ggml_backend_t backend) {
-    return backend != NULL && ggml_guid_matches(backend->guid, ggml_backend_vk_guid());
-}
+static ggml_backend_dev_t ggml_backend_vk_reg_get_device(ggml_backend_reg_t reg, size_t device) {
+    static std::vector devices;
 
-int ggml_backend_vk_get_device_count() {
-    return ggml_vk_get_device_count();
-}
+    static bool initialized = false;
 
-void ggml_backend_vk_get_device_description(int device, char * description, size_t description_size) {
-    ggml_vk_get_device_description(device, description, description_size);
-}
-
-void ggml_backend_vk_get_device_memory(int device, size_t * free, size_t * total) {
-    GGML_ASSERT(device < (int) vk_instance.device_indices.size());
-
-    vk::PhysicalDevice vkdev = vk_instance.instance.enumeratePhysicalDevices()[vk_instance.device_indices[device]];
-
-    vk::PhysicalDeviceMemoryProperties memprops = vkdev.getMemoryProperties();
-
-    for (const vk::MemoryHeap& heap : memprops.memoryHeaps) {
-        if (heap.flags & vk::MemoryHeapFlagBits::eDeviceLocal) {
-            *total = heap.size;
-            *free = heap.size;
-            break;
+    {
+        static std::mutex mutex;
+        std::lock_guard lock(mutex);
+        if (!initialized) {
+            for (size_t i = 0; i < ggml_backend_vk_get_device_count(); i++) {
+                ggml_backend_vk_device_context * ctx = new ggml_backend_vk_device_context;
+                char desc[256];
+                ggml_backend_vk_get_device_description(i, desc, sizeof(desc));
+                ctx->device = i;
+                ctx->name = GGML_VK_NAME + std::to_string(i);
+                ctx->description = desc;
+                devices.push_back(new ggml_backend_device {
+                    /* .iface   = */ ggml_backend_vk_device_i,
+                    /* .reg     = */ reg,
+                    /* .context = */ ctx,
+                });
+            }
+            initialized = true;
         }
     }
+
+    GGML_ASSERT(device < devices.size());
+    return devices[device];
+}
+
+static const struct ggml_backend_reg_i ggml_backend_vk_reg_i = {
+    /* .get_name         = */ ggml_backend_vk_reg_get_name,
+    /* .get_device_count = */ ggml_backend_vk_reg_get_device_count,
+    /* .get_device       = */ ggml_backend_vk_reg_get_device,
+    /* .get_proc_address = */ NULL,
+};
+
+ggml_backend_reg_t ggml_backend_vk_reg() {
+    static ggml_backend_reg reg = {
+        /* .iface   = */ ggml_backend_vk_reg_i,
+        /* .context = */ nullptr,
+    };
+
+    return ®
 }
 
 // Extension availability
diff --git a/src/llama.cpp b/src/llama.cpp
index c51b49c56..68479c6db 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -8,9 +8,7 @@
 #include "ggml-alloc.h"
 #include "ggml-backend.h"
 
-#if defined(GGML_USE_VULKAN)
-#  include "ggml-vulkan.h"
-#elif defined(GGML_USE_SYCL)
+#if defined(GGML_USE_SYCL)
 #  include "ggml-sycl.h"
 #elif defined(GGML_USE_KOMPUTE)
 #   include "ggml-kompute.h"
@@ -3418,8 +3416,6 @@ static int llama_get_device_count(const llama_model & model) {
 
 #if defined(GGML_USE_SYCL)
     count += ggml_backend_sycl_get_device_count();
-#elif defined(GGML_USE_VULKAN)
-    count += ggml_backend_vk_get_device_count();
 #elif defined(GGML_USE_CANN)
     count += ggml_backend_cann_get_device_count();
 #endif
@@ -3451,10 +3447,6 @@ static ggml_backend_buffer_type_t llama_default_buffer_type_cpu(const llama_mode
     }
 #elif defined(GGML_USE_CPU_HBM)
     buft = ggml_backend_cpu_hbm_buffer_type();
-#elif defined(GGML_USE_VULKAN)
-    if (host_buffer) {
-        buft = ggml_backend_vk_host_buffer_type();
-    }
 #endif
 
     if (buft == nullptr) {
@@ -3473,9 +3465,7 @@ static ggml_backend_buffer_type_t llama_default_buffer_type_offload(const llama_
     }
     device -= (int)model.devices.size();
 
-#if defined(GGML_USE_VULKAN)
-    buft = ggml_backend_vk_buffer_type(device);
-#elif defined(GGML_USE_SYCL)
+#if defined(GGML_USE_SYCL)
     buft = ggml_backend_sycl_buffer_type(device);
 #elif defined(GGML_USE_KOMPUTE)
     buft = ggml_backend_kompute_buffer_type(device);
@@ -3535,11 +3525,6 @@ static size_t llama_get_device_memory(const llama_model & model, int device) {
     size_t free;
     ggml_backend_sycl_get_device_memory(device, &free, &total);
     return free;
-#elif defined(GGML_USE_VULKAN)
-    size_t total;
-    size_t free;
-    ggml_backend_vk_get_device_memory(device, &free, &total);
-    return free;
 #elif defined(GGML_USE_CANN)
     size_t total;
     size_t free;
@@ -19095,8 +19080,7 @@ bool llama_supports_mlock(void) {
 }
 
 bool llama_supports_gpu_offload(void) {
-#if defined(GGML_USE_VULKAN) || \
-    defined(GGML_USE_SYCL) || defined(GGML_USE_KOMPUTE)
+#if defined(GGML_USE_SYCL) || defined(GGML_USE_KOMPUTE)
     // Defined when llama.cpp is compiled with support for offloading model layers to GPU.
     return true;
 #else
@@ -19227,8 +19211,13 @@ struct llama_model * llama_load_model_from_file(
 
             case GGML_BACKEND_DEVICE_TYPE_GPU:
             case GGML_BACKEND_DEVICE_TYPE_GPU_FULL:
+            {
+                size_t free, total; // NOLINT
+                ggml_backend_dev_memory(dev, &free, &total);
+                LLAMA_LOG_INFO("%s: using device %s (%s) - %zu MiB free\n", __func__, ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), free/1024/1024);
                 model->devices.push_back(dev);
                 break;
+            }
         }
     }
 
@@ -19423,32 +19412,7 @@ struct llama_context * llama_new_context_with_model(
             main_gpu -= (int)model->devices.size();
         }
 
-#if defined(GGML_USE_VULKAN)
-        if (model->split_mode == LLAMA_SPLIT_MODE_ROW) {
-            LLAMA_LOG_ERROR("%s: Row split not supported. Failed to initialize Vulkan backend\n", __func__);
-            llama_free(ctx);
-            return nullptr;
-        }
-        if (model->split_mode == LLAMA_SPLIT_MODE_NONE) {
-            ggml_backend_t backend = ggml_backend_vk_init(main_gpu);
-            if (backend == nullptr) {
-                LLAMA_LOG_ERROR("%s: failed to initialize Vulkan backend\n", __func__);
-                llama_free(ctx);
-                return nullptr;
-            }
-            ctx->backends.push_back(backend);
-        } else {
-            for (int device = 0; device < ggml_backend_vk_get_device_count(); ++device) {
-                ggml_backend_t backend = ggml_backend_vk_init(device);
-                if (backend == nullptr) {
-                    LLAMA_LOG_ERROR("%s: failed to initialize Vulkan%d backend\n", __func__, device);
-                    llama_free(ctx);
-                    return nullptr;
-                }
-                ctx->backends.push_back(backend);
-            }
-        }
-#elif defined(GGML_USE_SYCL)
+#if defined(GGML_USE_SYCL)
         // with split_mode LLAMA_SPLIT_MODE_NONE or LLAMA_SPLIT_MODE_ROW, only the main GPU backend is used
         if (model->split_mode == LLAMA_SPLIT_MODE_NONE || model->split_mode == LLAMA_SPLIT_MODE_ROW) {
             ggml_backend_t backend = ggml_backend_sycl_init(main_gpu);

From 3752217ed5a6a11864682fbf009bcb36afffd6bc Mon Sep 17 00:00:00 2001
From: Tim Wang 
Date: Thu, 17 Oct 2024 17:57:14 +1100
Subject: [PATCH 28/36] readme : update bindings list (#9918)

Co-authored-by: Tim Wang 
---
 README.md | 1 +
 1 file changed, 1 insertion(+)

diff --git a/README.md b/README.md
index 08fe8cc92..707904ddc 100644
--- a/README.md
+++ b/README.md
@@ -131,6 +131,7 @@ Typically finetunes of the base models below are supported as well.
 - PHP (API bindings and features built on top of llama.cpp): [distantmagic/resonance](https://github.com/distantmagic/resonance) [(more info)](https://github.com/ggerganov/llama.cpp/pull/6326)
 - Guile Scheme: [guile_llama_cpp](https://savannah.nongnu.org/projects/guile-llama-cpp)
 - Swift [srgtuszy/llama-cpp-swift](https://github.com/srgtuszy/llama-cpp-swift)
+- Swift [ShenghaiWang/SwiftLlama](https://github.com/ShenghaiWang/SwiftLlama)
 
 **UI:**
 

From 99bd4ac28c32cd17c0e337ff5601393b033dc5fc Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Thu, 17 Oct 2024 22:32:47 +0300
Subject: [PATCH 29/36] llama : infill sampling handle very long tokens (#9924)

* llama : infill sampling handle very long tokens

ggml-ci

* cont : better indices

ggml-ci
---
 include/llama.h        |  6 ------
 src/llama-sampling.cpp | 48 ++++++++++++++++++++++++++++++------------
 src/llama-vocab.cpp    | 17 ---------------
 src/llama.cpp          |  7 ------
 4 files changed, 35 insertions(+), 43 deletions(-)

diff --git a/include/llama.h b/include/llama.h
index 02bc7f087..1a13360c2 100644
--- a/include/llama.h
+++ b/include/llama.h
@@ -953,12 +953,6 @@ extern "C" {
                                int32_t   lstrip,
                                   bool   special);
 
-    // check if token0 is contained as a prefix in token1
-    LLAMA_API bool llama_token_is_prefix(
-              const struct llama_model * model,
-                           llama_token   token0,
-                           llama_token   token1);
-
     /// @details Convert the provided tokens into text (inverse of llama_tokenize()).
     /// @param text The char pointer must be large enough to hold the resulting text.
     /// @return Returns the number of chars/bytes on success, no more than text_len_max.
diff --git a/src/llama-sampling.cpp b/src/llama-sampling.cpp
index 2e6550682..bd750c40e 100644
--- a/src/llama-sampling.cpp
+++ b/src/llama-sampling.cpp
@@ -1745,6 +1745,9 @@ struct llama_sampler * llama_sampler_init_logit_bias(
 
 struct llama_sampler_infill {
     const struct llama_vocab * vocab;
+
+    std::vector buf0;
+    std::vector buf1;
 };
 
 static const char * llama_sampler_infill_name(const struct llama_sampler * /*smpl*/) {
@@ -1810,27 +1813,44 @@ static void llama_sampler_infill_apply(struct llama_sampler * smpl, llama_token_
     size_t n_combined = 0; GGML_UNUSED(n_combined);
 
     // combine tokens with common prefix
-    for (size_t i = 0; i < cur_p->size; ++i) {
-        for (size_t j = 0; j < cur_p->size; ++j) {
-            if (cur_p->data[i].logit == -INFINITY) {
+    for (size_t i0 = 0; i0 < cur_p->size; ++i0) {
+        for (size_t i1 = 0; i1 < cur_p->size; ++i1) {
+            if (cur_p->data[i0].logit == -INFINITY) {
                 break;
             }
 
-            if (i == j || cur_p->data[j].logit == -INFINITY) {
+            if (i0 == i1 || cur_p->data[i1].logit == -INFINITY) {
                 continue;
             }
 
-            if (llama_token_is_prefix_impl(*ctx->vocab, cur_p->data[i].id, cur_p->data[j].id)) {
-                if (cur_p->data[i].p >  cur_p->data[j].p) {
-                    cur_p->data[i].p += cur_p->data[j].p;
-                    cur_p->data[j].logit = -INFINITY;
-                    cur_p->data[j].p     = 0.0f;
-                } else {
-                    cur_p->data[j].p += cur_p->data[i].p;
-                    cur_p->data[i].logit = -INFINITY;
-                    cur_p->data[i].p     = 0.0f;
+            int len0 = llama_token_to_piece_impl(*ctx->vocab, cur_p->data[i0].id, ctx->buf0.data(), ctx->buf0.size(), 0, false);
+            if (len0 < 0) {
+                ctx->buf0.resize(len0);
+                len0 = llama_token_to_piece_impl(*ctx->vocab, cur_p->data[i0].id, ctx->buf0.data(), ctx->buf0.size(), 0, false);
+                assert(len0 > 0);
+            }
+
+            int len1 = llama_token_to_piece_impl(*ctx->vocab, cur_p->data[i1].id, ctx->buf1.data(), ctx->buf1.size(), 0, false);
+            if (len1 < 0) {
+                ctx->buf1.resize(len1);
+                len1 = llama_token_to_piece_impl(*ctx->vocab, cur_p->data[i1].id, ctx->buf1.data(), ctx->buf1.size(), 0, false);
+                assert(len1 > 0);
+            }
+
+            // token i0 is a prefix of token i1
+            if (len0 > 0 && len0 <= len1 && memcmp(ctx->buf0.data(), ctx->buf1.data(), len0) == 0) {
+                int dst = i0;
+                int src = i1;
+
+                // merge into the token with higher probability
+                if (cur_p->data[i1].p > cur_p->data[i0].p) {
+                    std::swap(dst, src);
                 }
 
+                cur_p->data[dst].p += cur_p->data[src].p;
+                cur_p->data[src].logit = -INFINITY;
+                cur_p->data[src].p     = 0.0f;
+
                 n_combined++;
             }
         }
@@ -1936,6 +1956,8 @@ struct llama_sampler * llama_sampler_init_infill_impl(
         /* .iface = */ &llama_sampler_infill_i,
         /* .ctx   = */ new llama_sampler_infill {
             /* .vocab = */ &vocab,
+            /* .buf0 = */ std::vector(512),
+            /* .buf1 = */ std::vector(512),
         },
     };
 }
diff --git a/src/llama-vocab.cpp b/src/llama-vocab.cpp
index 57d56a3d3..0a49ddbe3 100644
--- a/src/llama-vocab.cpp
+++ b/src/llama-vocab.cpp
@@ -1858,23 +1858,6 @@ int32_t llama_token_to_piece_impl(const struct llama_vocab & vocab, llama_token
     return 0;
 }
 
-bool llama_token_is_prefix_impl(
-        const struct llama_vocab & vocab,
-                     llama_token   token0,
-                     llama_token   token1) {
-    char text_buf_0[128];
-    char text_buf_1[128];
-
-    const int32_t len0 = llama_token_to_piece_impl(vocab, token0, text_buf_0, sizeof(text_buf_0) - 1, 0, false);
-    const int32_t len1 = llama_token_to_piece_impl(vocab, token1, text_buf_1, sizeof(text_buf_1) - 1, 0, false);
-
-    if (len0 <= 0 || len1 <= 0) {
-        return false;
-    }
-
-    return len0 <= len1 && memcmp(text_buf_0, text_buf_1, len0) == 0;
-}
-
 int32_t llama_detokenize_impl(
         const struct llama_vocab & vocab,
                const llama_token * tokens,
diff --git a/src/llama.cpp b/src/llama.cpp
index 68479c6db..d8e2b006c 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -21466,13 +21466,6 @@ int32_t llama_token_to_piece(
     return llama_token_to_piece_impl(model->vocab, token, buf, length, lstrip, special);
 }
 
-bool llama_token_is_prefix(
-    const struct llama_model * model,
-                 llama_token   token0,
-                 llama_token   token1) {
-    return llama_token_is_prefix_impl(model->vocab, token0, token1);
-}
-
 int32_t llama_detokenize(
     const struct llama_model * model,
            const llama_token * tokens,

From 9f45fc1e9950a496febc575cdd196cd5cad000cc Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Thu, 17 Oct 2024 23:26:32 +0300
Subject: [PATCH 30/36] llama : change warning to debug log

---
 src/llama.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/llama.cpp b/src/llama.cpp
index d8e2b006c..ffaa6f789 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -6735,9 +6735,9 @@ static void llm_load_vocab(
                     vocab.id_to_token[t.second].attr = LLAMA_TOKEN_ATTR_CONTROL;
                 }
             } else {
-                // token is control, but not marked as EOG -> print a warning
+                // token is control, but not marked as EOG -> print a debug log
                 if (vocab.id_to_token[t.second].attr & LLAMA_TOKEN_ATTR_CONTROL && vocab.special_eog_ids.count(t.second) == 0) {
-                    LLAMA_LOG_WARN("%s: control token: %6d '%s' is not marked as EOG\n",
+                    LLAMA_LOG_DEBUG("%s: control token: %6d '%s' is not marked as EOG\n",
                             __func__, t.second, t.first.c_str());
                 }
             }

From 17bb9280807cfbb6611b853aa1ef05114bd9efe9 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Thu, 17 Oct 2024 23:43:05 +0300
Subject: [PATCH 31/36] readme : remove --memory-f32 references (#9925)

---
 examples/main/README.md    | 4 ----
 scripts/run-with-preset.py | 6 +++---
 2 files changed, 3 insertions(+), 7 deletions(-)

diff --git a/examples/main/README.md b/examples/main/README.md
index 620934dad..7e192b9f2 100644
--- a/examples/main/README.md
+++ b/examples/main/README.md
@@ -297,10 +297,6 @@ These options help improve the performance and memory usage of the LLaMA models.
 
  These flags attempt optimizations that help on some systems with non-uniform memory access. This currently consists of one of the above strategies, and disabling prefetch and readahead for mmap. The latter causes mapped pages to be faulted in on first access instead of all at once, and in combination with pinning threads to NUMA nodes, more of the pages end up on the NUMA node where they are used. Note that if the model is already in the system page cache, for example because of a previous run without this option, this will have little effect unless you drop the page cache first. This can be done by rebooting the system or on Linux by writing '3' to '/proc/sys/vm/drop_caches' as root.
 
-### Memory Float 32
-
--   `--memory-f32`: Use 32-bit floats instead of 16-bit floats for memory key+value. This doubles the context memory requirement and cached prompt file size but does not appear to increase generation quality in a measurable way. Not recommended.
-
 ### Batch Size
 
 -   `-b N, --batch-size N`: Set the batch size for prompt processing (default: `2048`). This large batch size benefits users who have BLAS installed and enabled it during the build. If you don't have BLAS enabled ("BLAS=0"), you can use a smaller number, such as 8, to see the prompt progress as it's evaluated in some situations.
diff --git a/scripts/run-with-preset.py b/scripts/run-with-preset.py
index ee21eab37..47cacb432 100755
--- a/scripts/run-with-preset.py
+++ b/scripts/run-with-preset.py
@@ -15,7 +15,7 @@ CLI_ARGS_LLAMA_CLI_PERPLEXITY = [
     "export", "file", "frequency-penalty", "grammar", "grammar-file", "hellaswag",
     "hellaswag-tasks", "ignore-eos", "in-prefix", "in-prefix-bos", "in-suffix",
     "interactive", "interactive-first", "keep", "logdir", "logit-bias", "lora", "lora-base",
-    "low-vram", "main-gpu", "memory-f32", "mirostat", "mirostat-ent", "mirostat-lr", "mlock",
+    "low-vram", "main-gpu", "mirostat", "mirostat-ent", "mirostat-lr", "mlock",
     "model", "multiline-input", "n-gpu-layers", "n-predict", "no-mmap", "no-mul-mat-q",
     "np-penalize-nl", "numa", "ppl-output-type", "ppl-stride", "presence-penalty", "prompt",
     "prompt-cache", "prompt-cache-all", "prompt-cache-ro", "repeat-last-n",
@@ -25,12 +25,12 @@ CLI_ARGS_LLAMA_CLI_PERPLEXITY = [
 ]
 
 CLI_ARGS_LLAMA_BENCH = [
-    "batch-size", "memory-f32", "low-vram", "model", "mul-mat-q", "n-gen", "n-gpu-layers",
+    "batch-size", "low-vram", "model", "mul-mat-q", "n-gen", "n-gpu-layers",
     "n-prompt", "output", "repetitions", "tensor-split", "threads", "verbose"
 ]
 
 CLI_ARGS_LLAMA_SERVER = [
-    "alias", "batch-size", "ctx-size", "embedding", "host", "memory-f32", "lora", "lora-base",
+    "alias", "batch-size", "ctx-size", "embedding", "host", "lora", "lora-base",
     "low-vram", "main-gpu", "mlock", "model", "n-gpu-layers", "n-probs", "no-mmap", "no-mul-mat-q",
     "numa", "path", "port", "rope-freq-base", "timeout", "rope-freq-scale", "tensor-split",
     "threads", "verbose"

From 6f55bccbb8835d42147add4ee48807450f5ff535 Mon Sep 17 00:00:00 2001
From: Daniel Bevenius 
Date: Fri, 18 Oct 2024 01:41:51 +0200
Subject: [PATCH 32/36] llama : rename batch_all to batch (#8881)

This commit addresses the TODO in the code to rename the `batch_all`
parameter to `batch` in `llama_decode_internal`.
---
 src/llama.cpp | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

diff --git a/src/llama.cpp b/src/llama.cpp
index ffaa6f789..dcb015d12 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -17134,10 +17134,10 @@ static void llama_graph_compute(
 //
 static int llama_decode_internal(
          llama_context & lctx,
-           llama_batch   batch_all) { // TODO: rename back to batch
+           llama_batch   batch) {
 
     lctx.is_encoding = false;
-    const uint32_t n_tokens_all = batch_all.n_tokens;
+    const uint32_t n_tokens_all = batch.n_tokens;
 
     if (n_tokens_all == 0) {
         LLAMA_LOG_ERROR("%s: n_tokens == 0\n", __func__);
@@ -17148,12 +17148,12 @@ static int llama_decode_internal(
     const auto & hparams = model.hparams;
     const auto & cparams = lctx.cparams;
 
-    GGML_ASSERT((!batch_all.token && batch_all.embd) || (batch_all.token && !batch_all.embd)); // NOLINT
+    GGML_ASSERT((!batch.token && batch.embd) || (batch.token && !batch.embd)); // NOLINT
 
-    if (batch_all.token) {
+    if (batch.token) {
         for (uint32_t i = 0; i < n_tokens_all; ++i) {
-            if (batch_all.token[i] < 0 || (uint32_t)batch_all.token[i] >= model.vocab.n_vocab) {
-                LLAMA_LOG_ERROR("%s: invalid token[%d] = %d\n", __func__, i, batch_all.token[i]);
+            if (batch.token[i] < 0 || (uint32_t)batch.token[i] >= model.vocab.n_vocab) {
+                LLAMA_LOG_ERROR("%s: invalid token[%d] = %d\n", __func__, i, batch.token[i]);
                 return -1;
             }
         }
@@ -17184,9 +17184,9 @@ static int llama_decode_internal(
     lctx.embd_seq.clear();
 
     // count outputs
-    if (batch_all.logits && !embd_pooled) {
+    if (batch.logits && !embd_pooled) {
         for (uint32_t i = 0; i < n_tokens_all; ++i) {
-            n_outputs += batch_all.logits[i] != 0;
+            n_outputs += batch.logits[i] != 0;
         }
     } else if (lctx.logits_all || embd_pooled) {
         n_outputs = n_tokens_all;
@@ -17195,7 +17195,7 @@ static int llama_decode_internal(
         n_outputs = 1;
     }
 
-    lctx.sbatch.from_batch(batch_all, n_embd,
+    lctx.sbatch.from_batch(batch, n_embd,
         /* simple_split */ !kv_self.recurrent,
         /* logits_all   */ n_outputs == n_tokens_all);
 

From 8901755ba328643c9ab071c20e1939ea52951a0e Mon Sep 17 00:00:00 2001
From: Georgi Gerganov 
Date: Fri, 18 Oct 2024 07:32:19 +0300
Subject: [PATCH 33/36] server : add n_indent parameter for line indentation
 requirement (#9929)

ggml-ci
---
 examples/server/README.md  |  2 ++
 examples/server/server.cpp | 54 +++++++++++++++++++++++++++++++++-----
 2 files changed, 49 insertions(+), 7 deletions(-)

diff --git a/examples/server/README.md b/examples/server/README.md
index fcdb02afd..09f1aa249 100644
--- a/examples/server/README.md
+++ b/examples/server/README.md
@@ -333,6 +333,8 @@ node index.js
 
     `n_predict`: Set the maximum number of tokens to predict when generating text. **Note:** May exceed the set limit slightly if the last token is a partial multibyte character. When 0, no tokens will be generated but the prompt is evaluated into the cache. Default: `-1`, where `-1` is infinity.
 
+    `n_indent`: Specify the minimum line indentation for the generated text in number of whitespace characters. Useful for code completion tasks. Default: `0`
+
     `n_keep`: Specify the number of tokens from the prompt to retain when the context size is exceeded and tokens need to be discarded. The number excludes the BOS token.
     By default, this value is set to `0`, meaning no tokens are kept. Use `-1` to retain all tokens from the prompt.
 
diff --git a/examples/server/server.cpp b/examples/server/server.cpp
index b5e63384c..8fd443878 100644
--- a/examples/server/server.cpp
+++ b/examples/server/server.cpp
@@ -131,6 +131,7 @@ struct slot_params {
     int32_t n_keep    =  0; // number of tokens to keep from initial prompt
     int32_t n_discard =  0; // number of tokens after n_keep that may be discarded when shifting context, 0 defaults to half
     int32_t n_predict = -1; // new tokens to predict
+    int32_t n_indent  =  0; // mininum line indentation for the generated text in number of whitespace characters
 
     int64_t t_max_prompt_ms  = -1; // TODO: implement
     int64_t t_max_predict_ms = -1; // if positive, limit the generation phase to this time limit
@@ -173,6 +174,8 @@ struct server_slot {
     std::vector prompt_tokens;
     std::vector extra_tokens;
 
+    size_t last_nl_pos = 0;
+
     std::string generated_text;
     std::vector cache_tokens;
     std::vector generated_token_probs;
@@ -215,6 +218,7 @@ struct server_slot {
         SLT_DBG(*this, "%s", "\n");
 
         n_prompt_tokens    = 0;
+        last_nl_pos        = 0;
         generated_text     = "";
         has_new_line       = false;
         truncated          = false;
@@ -860,6 +864,7 @@ struct server_context {
         slot.params.stream             = json_value(data, "stream",            false);
         slot.params.cache_prompt       = json_value(data, "cache_prompt",      false);
         slot.params.n_predict          = json_value(data, "n_predict",         json_value(data, "max_tokens", default_params.n_predict));
+        slot.params.n_indent           = json_value(data, "n_indent",          default_params.n_indent);
         slot.sparams.top_k             = json_value(data, "top_k",             default_sparams.top_k);
         slot.sparams.top_p             = json_value(data, "top_p",             default_sparams.top_p);
         slot.sparams.min_p             = json_value(data, "min_p",             default_sparams.min_p);
@@ -878,7 +883,7 @@ struct server_context {
         slot.sparams.mirostat_tau      = json_value(data, "mirostat_tau",      default_sparams.mirostat_tau);
         slot.sparams.mirostat_eta      = json_value(data, "mirostat_eta",      default_sparams.mirostat_eta);
         slot.sparams.penalize_nl       = json_value(data, "penalize_nl",       default_sparams.penalize_nl);
-        slot.params.n_keep             = json_value(data, "n_keep",            slot.params.n_keep);
+        slot.params.n_keep             = json_value(data, "n_keep",            default_params.n_keep);
         slot.params.n_discard          = json_value(data, "n_discard",         default_params.n_discard);
         slot.sparams.seed              = json_value(data, "seed",              default_sparams.seed);
         slot.sparams.n_probs           = json_value(data, "n_probs",           default_sparams.n_probs);
@@ -1129,13 +1134,48 @@ struct server_context {
             SLT_DBG(slot, "stopped by limit, n_decoded = %d, n_predict = %d\n", slot.n_decoded, slot.params.n_predict);
         }
 
-        // if we have already seen a new line, we stop after a certain time limit
-        if (slot.has_new_line && slot.params.t_max_predict_ms > 0 &&
-            (ggml_time_us() - slot.t_start_generation > 1000.0f*slot.params.t_max_predict_ms)) {
-            slot.stopped_limit  = true;
-            slot.has_next_token = false;
+        if (slot.has_new_line) {
+            // if we have already seen a new line, we stop after a certain time limit
+            if (slot.params.t_max_predict_ms > 0 && (ggml_time_us() - slot.t_start_generation > 1000.0f*slot.params.t_max_predict_ms)) {
+                slot.stopped_limit  = true;
+                slot.has_next_token = false;
 
-            SLT_DBG(slot, "stopped by time limit, n_decoded = %d, t_max_predict_ms = %d ms\n", slot.n_decoded, (int) slot.params.t_max_predict_ms);
+                SLT_DBG(slot, "stopped by time limit, n_decoded = %d, t_max_predict_ms = %d ms\n", slot.n_decoded, (int) slot.params.t_max_predict_ms);
+            }
+
+            // require that each new line has a whitespace prefix (i.e. indentation) of at least slot.params.n_indent
+            if (slot.params.n_indent > 0) {
+                // check the current indentation
+                // TODO: improve by not doing it more than once for each new line
+                if (slot.last_nl_pos > 0) {
+                    size_t pos = slot.last_nl_pos;
+
+                    int n_indent = 0;
+                    while (pos < slot.generated_text.size() && (slot.generated_text[pos] == ' ' || slot.generated_text[pos] == '\t')) {
+                        n_indent++;
+                        pos++;
+                    }
+
+                    if (pos < slot.generated_text.size() && n_indent < slot.params.n_indent) {
+                        slot.stopped_limit  = true;
+                        slot.has_next_token = false;
+
+                        // cut the last line
+                        slot.generated_text.erase(pos, std::string::npos);
+
+                        SLT_DBG(slot, "stopped by indentation limit, n_decoded = %d, n_indent = %d\n", slot.n_decoded, n_indent);
+                    }
+                }
+
+                // find the next new line
+                {
+                    const size_t pos = slot.generated_text.find('\n', slot.last_nl_pos);
+
+                    if (pos != std::string::npos) {
+                        slot.last_nl_pos = pos + 1;
+                    }
+                }
+            }
         }
 
         // check if there is a new line in the generated text

From 60ce97c9d809f4b040e90b597468b839df5728d0 Mon Sep 17 00:00:00 2001
From: Ma Mingfei 
Date: Fri, 18 Oct 2024 13:34:36 +0800
Subject: [PATCH 34/36] add amx kernel for gemm (#8998)

add intel amx isa detection

add vnni kernel for gemv cases

add vnni and amx kernel support for block_q8_0

code cleanup

fix packing B issue

enable openmp

fine tune amx kernel

switch to aten parallel pattern

add error message for nested parallelism

code cleanup

add f16 support in ggml-amx

add amx kernels for QK_K quant formats: Q4_K, Q5_K, Q6_K and IQ4_XS

update CMakeList

update README

fix some compilation warning

fix compiler warning when amx is not enabled

minor change

ggml-ci

move ggml_amx_init from ggml.c to ggml-amx/mmq.cpp

ggml-ci

update CMakeLists with -mamx-tile, -mamx-int8 and -mamx-bf16

ggml-ci

add amx as an ggml-backend

update header file, the old path for immintrin.h has changed to ggml-cpu-impl.h

minor change

update CMakeLists.txt

minor change

apply weight prepacking in set_tensor method in ggml-backend

fix compile error

ggml-ci

minor change

ggml-ci

update CMakeLists.txt

ggml-ci

add march dependency

minor change

ggml-ci

change ggml_backend_buffer_is_host to return false for amx backend

ggml-ci

fix supports_op

use device reg for AMX backend

ggml-ci

minor change

ggml-ci

minor change

fix rebase

set .buffer_from_host_ptr to be false for AMX backend
---
 CMakeLists.txt             |    4 +
 Makefile                   |   24 +-
 README.md                  |    2 +-
 ggml/CMakeLists.txt        |    4 +
 ggml/include/ggml-amx.h    |   25 +
 ggml/include/ggml.h        |    1 +
 ggml/src/CMakeLists.txt    |   42 +
 ggml/src/ggml-amx.cpp      |  453 +++++++
 ggml/src/ggml-amx/common.h |   93 ++
 ggml/src/ggml-amx/mmq.cpp  | 2509 ++++++++++++++++++++++++++++++++++++
 ggml/src/ggml-amx/mmq.h    |   17 +
 ggml/src/ggml-backend.cpp  |   12 +-
 ggml/src/ggml.c            |    8 +
 src/llama.cpp              |   17 +
 14 files changed, 3204 insertions(+), 7 deletions(-)
 create mode 100644 ggml/include/ggml-amx.h
 create mode 100644 ggml/src/ggml-amx.cpp
 create mode 100644 ggml/src/ggml-amx/common.h
 create mode 100644 ggml/src/ggml-amx/mmq.cpp
 create mode 100644 ggml/src/ggml-amx/mmq.h

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 64a335378..ef0932a7b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -88,6 +88,10 @@ if (NOT DEFINED GGML_LLAMAFILE)
     set(GGML_LLAMAFILE_DEFAULT ON)
 endif()
 
+if (NOT DEFINED GGML_AMX)
+    set(GGML_AMX ON)
+endif()
+
 if (NOT DEFINED GGML_CUDA_GRAPHS)
     set(GGML_CUDA_GRAPHS_DEFAULT ON)
 endif()
diff --git a/Makefile b/Makefile
index 2793978c3..719f45d16 100644
--- a/Makefile
+++ b/Makefile
@@ -93,11 +93,6 @@ GGML_METAL := 1
 DEPRECATE_WARNING := 1
 endif
 
-ifdef LLAMA_OPENMP
-GGML_OPENMP := 1
-DEPRECATE_WARNING := 1
-endif
-
 ifdef LLAMA_RPC
 GGML_RPC := 1
 DEPRECATE_WARNING := 1
@@ -584,6 +579,11 @@ ifndef GGML_NO_LLAMAFILE
 	OBJ_GGML    += ggml/src/llamafile/sgemm.o
 endif
 
+ifndef GGML_NO_AMX
+	MK_CPPFLAGS += -DGGML_USE_AMX
+	OBJ_GGML    += ggml/src/ggml-amx.o ggml/src/ggml-amx/mmq.o
+endif
+
 ifdef GGML_RPC
 	MK_CPPFLAGS += -DGGML_USE_RPC
 	OBJ_GGML    += ggml/src/ggml-rpc.o
@@ -1087,6 +1087,19 @@ ggml/src/llamafile/sgemm.o: \
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 endif # GGML_NO_LLAMAFILE
 
+ifndef GGML_NO_AMX
+ggml/src/ggml-amx.o: \
+	ggml/src/ggml-amx.cpp \
+	ggml/include/ggml-amx.h
+	$(CXX) $(CXXFLAGS) -c $< -o $@
+
+ggml/src/ggml-amx/mmq.o: \
+	ggml/src/ggml-amx/mmq.cpp \
+	ggml/src/ggml-amx/mmq.h \
+	ggml/include/ggml.h
+	$(CXX) $(CXXFLAGS) -c $< -o $@
+endif
+
 ifdef GGML_RPC
 ggml/src/ggml-rpc.o: \
 	ggml/src/ggml-rpc.cpp \
@@ -1238,6 +1251,7 @@ clean:
 	rm -vrf ggml/src/ggml-metal-embed.metal
 	rm -vrf ggml/src/ggml-cuda/*.o
 	rm -vrf ggml/src/ggml-cuda/template-instances/*.o
+	rm -vrf ggml/src/ggml-amx/*.o
 	rm -rvf $(BUILD_TARGETS)
 	rm -rvf $(TEST_TARGETS)
 	rm -f vulkan-shaders-gen ggml/src/ggml-vulkan-shaders.hpp ggml/src/ggml-vulkan-shaders.cpp
diff --git a/README.md b/README.md
index 707904ddc..1088b3338 100644
--- a/README.md
+++ b/README.md
@@ -29,7 +29,7 @@ variety of hardware - locally and in the cloud.
 
 - Plain C/C++ implementation without any dependencies
 - Apple silicon is a first-class citizen - optimized via ARM NEON, Accelerate and Metal frameworks
-- AVX, AVX2 and AVX512 support for x86 architectures
+- AVX, AVX2, AVX512 and AMX support for x86 architectures
 - 1.5-bit, 2-bit, 3-bit, 4-bit, 5-bit, 6-bit, and 8-bit integer quantization for faster inference and reduced memory use
 - Custom CUDA kernels for running LLMs on NVIDIA GPUs (support for AMD GPUs via HIP and Moore Threads MTT GPUs via MUSA)
 - Vulkan and SYCL backend support
diff --git a/ggml/CMakeLists.txt b/ggml/CMakeLists.txt
index 89fdf9d1c..cfa6e3f70 100644
--- a/ggml/CMakeLists.txt
+++ b/ggml/CMakeLists.txt
@@ -99,6 +99,9 @@ option(GGML_AVX512      "ggml: enable AVX512"           OFF)
 option(GGML_AVX512_VBMI "ggml: enable AVX512-VBMI"      OFF)
 option(GGML_AVX512_VNNI "ggml: enable AVX512-VNNI"      OFF)
 option(GGML_AVX512_BF16 "ggml: enable AVX512-BF16"      OFF)
+option(GGML_AMX_TILE    "ggml: enable AMX-TILE"         OFF)
+option(GGML_AMX_INT8    "ggml: enable AMX-INT8"         OFF)
+option(GGML_AMX_BF16    "ggml: enable AMX-BF16"         OFF)
 option(GGML_FMA         "ggml: enable FMA"              ${INS_ENB})
 if (NOT MSVC)
     option(GGML_F16C    "ggml: enable F16C"             ${INS_ENB}) # in MSVC F16C is implied with AVX2/AVX512
@@ -158,6 +161,7 @@ set   (GGML_METAL_MACOSX_VERSION_MIN "" CACHE STRING
 set   (GGML_METAL_STD "" CACHE STRING       "ggml: metal standard version (-std flag)")
 option(GGML_OPENMP                          "ggml: use OpenMP"                                ON)
 option(GGML_RPC                             "ggml: use RPC"                                   OFF)
+option(GGML_AMX                             "ggml: use AMX"                                   OFF)
 option(GGML_SYCL                            "ggml: use SYCL"                                  OFF)
 option(GGML_SYCL_F16                        "ggml: use 16 bit floats for sycl calculations"   OFF)
 set   (GGML_SYCL_TARGET "INTEL" CACHE STRING
diff --git a/ggml/include/ggml-amx.h b/ggml/include/ggml-amx.h
new file mode 100644
index 000000000..22b3f70f4
--- /dev/null
+++ b/ggml/include/ggml-amx.h
@@ -0,0 +1,25 @@
+#pragma once
+
+#include "ggml.h"
+#include "ggml-backend.h"
+
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+// buffer_type API
+GGML_API ggml_backend_buffer_type_t ggml_backend_amx_buffer_type(void);
+
+GGML_API bool ggml_backend_is_amx(ggml_backend_t backend);
+
+// backend API
+GGML_API ggml_backend_t ggml_backend_amx_init(void);
+
+GGML_API void ggml_backend_amx_set_n_threads(ggml_backend_t backend_amx, int n_threads);
+
+GGML_API ggml_backend_reg_t ggml_backend_amx_reg(void);
+
+#ifdef  __cplusplus
+}
+#endif
diff --git a/ggml/include/ggml.h b/ggml/include/ggml.h
index 4508da4fb..de3c706fc 100644
--- a/ggml/include/ggml.h
+++ b/ggml/include/ggml.h
@@ -2488,6 +2488,7 @@ extern "C" {
     GGML_API int ggml_cpu_has_avx512_vbmi(void);
     GGML_API int ggml_cpu_has_avx512_vnni(void);
     GGML_API int ggml_cpu_has_avx512_bf16(void);
+    GGML_API int ggml_cpu_has_amx_int8   (void);
     GGML_API int ggml_cpu_has_fma        (void);
     GGML_API int ggml_cpu_has_neon       (void);
     GGML_API int ggml_cpu_has_sve        (void);
diff --git a/ggml/src/CMakeLists.txt b/ggml/src/CMakeLists.txt
index 676f85a36..aa405e4d0 100644
--- a/ggml/src/CMakeLists.txt
+++ b/ggml/src/CMakeLists.txt
@@ -267,6 +267,26 @@ if (GGML_LLAMAFILE)
     set(GGML_SOURCES_LLAMAFILE llamafile/sgemm.cpp)
 endif()
 
+if (GGML_AMX)
+    if (CMAKE_COMPILER_IS_GNUCC AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 11.0)
+    else()
+        set(GGML_AMX OFF)
+        message(WARNING "AMX requires gcc version > 11.0. Turning off GGML_AMX.")
+    endif()
+
+    if (GGML_AMX)
+        message(STATUS "Using AMX")
+
+        list(APPEND GGML_CDEF_PUBLIC GGML_USE_AMX)
+
+        file(GLOB   GGML_HEADERS_AMX "ggml-amx/*.h")
+        list(APPEND GGML_HEADERS_AMX "../include/ggml-amx.h")
+
+        file(GLOB   GGML_SOURCES_AMX "ggml-amx/*.cpp")
+        list(APPEND GGML_SOURCES_AMX "ggml-amx.cpp")
+    endif()
+endif()
+
 if (GGML_CUDA)
     cmake_minimum_required(VERSION 3.18)  # for CMAKE_CUDA_ARCHITECTURES
 
@@ -1180,6 +1200,18 @@ elseif (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR CMAKE_GENERATOR_PLATFORM_LW
                 add_compile_definitions($<$:__AVX512BF16__>)
                 add_compile_definitions($<$:__AVX512BF16__>)
             endif()
+            if (GGML_AMX_TILE)
+                add_compile_definitions($<$:__AMX_TILE__>)
+                add_compile_definitions($<$:__AMX_TILE__>)
+            endif()
+            if (GGML_AMX_INT8)
+                add_compile_definitions($<$:__AMX_INT8__>)
+                add_compile_definitions($<$:__AMX_INT8__>)
+            endif()
+            if (GGML_AMX_BF16)
+                add_compile_definitions($<$:__AMX_BF16__>)
+                add_compile_definitions($<$:__AMX_BF16__>)
+            endif()
         elseif (GGML_AVX2)
             list(APPEND ARCH_FLAGS /arch:AVX2)
         elseif (GGML_AVX)
@@ -1215,6 +1247,15 @@ elseif (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR CMAKE_GENERATOR_PLATFORM_LW
         if (GGML_AVX512_BF16)
             list(APPEND ARCH_FLAGS -mavx512bf16)
         endif()
+        if (GGML_AMX_TILE)
+            list(APPEND ARCH_FLAGS -mamx-tile)
+        endif()
+        if (GGML_AMX_INT8)
+            list(APPEND ARCH_FLAGS -mamx-int8)
+        endif()
+        if (GGML_AMX_BF16)
+            list(APPEND ARCH_FLAGS -mamx-bf16)
+        endif()
     endif()
 elseif (${CMAKE_SYSTEM_PROCESSOR} MATCHES "ppc64")
     message(STATUS "PowerPC detected")
@@ -1340,6 +1381,7 @@ add_library(ggml
             ${GGML_SOURCES_ROCM}      ${GGML_HEADERS_ROCM}
             ${GGML_SOURCES_BLAS}      ${GGML_HEADERS_BLAS}
             ${GGML_SOURCES_LLAMAFILE} ${GGML_HEADERS_LLAMAFILE}
+            ${GGML_SOURCES_AMX}       ${GGML_HEADERS_AMX}
             ${GGML_SOURCES_CANN}      ${GGML_HEADERS_CANN}
             ggml-aarch64.c            ggml-aarch64.h
             )
diff --git a/ggml/src/ggml-amx.cpp b/ggml/src/ggml-amx.cpp
new file mode 100644
index 000000000..ac6ec2342
--- /dev/null
+++ b/ggml/src/ggml-amx.cpp
@@ -0,0 +1,453 @@
+#include "ggml-amx.h"
+#include "ggml-amx/common.h"
+#include "ggml-amx/mmq.h"
+#include "ggml-backend-impl.h"
+#include "ggml-impl.h"
+
+#if defined(__gnu_linux__)
+#include 
+#include 
+#endif
+
+#include 
+#include 
+#include 
+
+#if defined(__AMX_INT8__)
+
+// AMX buffer interface
+static const char * ggml_backend_amx_buffer_get_name(ggml_backend_buffer_t buffer) {
+    return "AMX";
+
+    GGML_UNUSED(buffer);
+}
+
+static void ggml_backend_amx_buffer_free_buffer(ggml_backend_buffer_t buffer) {
+    free(buffer->context);
+}
+
+static void * ggml_backend_amx_buffer_get_base(ggml_backend_buffer_t buffer) {
+    return (void *)(buffer->context);
+}
+
+static void ggml_backend_amx_buffer_memset_tensor(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, uint8_t value, size_t offset, size_t size) {
+    memset((char *)tensor->data + offset, value, size);
+
+    GGML_UNUSED(buffer);
+}
+
+static void ggml_backend_amx_buffer_set_tensor(ggml_backend_buffer_t buffer, struct ggml_tensor * tensor, const void * data, size_t offset, size_t size) {
+    if (qtype_has_amx_kernels(tensor->type)) {
+        ggml_backend_amx_convert_weight(tensor, data, offset, size);
+    } else {
+        memcpy((char *)tensor->data + offset, data, size);
+    }
+
+    GGML_UNUSED(buffer);
+}
+
+static void ggml_backend_amx_buffer_get_tensor(ggml_backend_buffer_t buffer, const struct ggml_tensor * tensor, void * data, size_t offset, size_t size) {
+    GGML_ASSERT(!qtype_has_amx_kernels(tensor->type));
+    memcpy(data, (const char *)tensor->data + offset, size);
+
+    GGML_UNUSED(buffer);
+}
+
+static bool ggml_backend_amx_buffer_cpy_tensor(ggml_backend_buffer_t buffer, const struct ggml_tensor * src, struct ggml_tensor * dst) {
+    if (ggml_backend_buffer_is_host(src->buffer)) {
+        if (qtype_has_amx_kernels(src->type)) {
+            ggml_backend_amx_convert_weight(dst, src->data, 0, ggml_backend_amx_get_alloc_size(dst));
+        } else {
+            memcpy(dst->data, src->data, ggml_nbytes(src));
+        }
+        return true;
+    }
+    return false;
+
+    GGML_UNUSED(buffer);
+}
+
+static void ggml_backend_amx_buffer_clear(ggml_backend_buffer_t buffer, uint8_t value) {
+    memset(buffer->context, value, buffer->size);
+}
+
+static ggml_backend_buffer_i ggml_backend_amx_buffer_interface = {
+    /* .get_name        = */ ggml_backend_amx_buffer_get_name,
+    /* .free_buffer     = */ ggml_backend_amx_buffer_free_buffer,
+    /* .get_base        = */ ggml_backend_amx_buffer_get_base,
+    /* .init_tensor     = */ NULL, // no initialization required
+    /* .memset_tensor   = */ ggml_backend_amx_buffer_memset_tensor,
+    /* .set_tensor      = */ ggml_backend_amx_buffer_set_tensor,
+    /* .get_tensor      = */ ggml_backend_amx_buffer_get_tensor,
+    /* .cpy_tensor      = */ ggml_backend_amx_buffer_cpy_tensor,
+    /* .clear           = */ ggml_backend_amx_buffer_clear,
+    /* .reset           = */ NULL,
+};
+
+static const char * ggml_backend_amx_buffer_type_get_name(ggml_backend_buffer_type_t buft) {
+    return "AMX";
+
+    GGML_UNUSED(buft);
+}
+
+static ggml_backend_buffer_t ggml_backend_amx_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
+    void * data = aligned_alloc(TENSOR_ALIGNMENT, size);
+    if (data == NULL) {
+        fprintf(stderr, "%s: failed to allocate buffer of size %zu\n", __func__, size);
+        return NULL;
+    }
+
+    return ggml_backend_buffer_init(buft, ggml_backend_amx_buffer_interface, data, size);
+}
+
+static size_t ggml_backend_amx_buffer_type_get_alignment(ggml_backend_buffer_type_t buft) {
+    return TENSOR_ALIGNMENT;
+
+    GGML_UNUSED(buft);
+}
+
+static size_t ggml_backend_amx_buffer_type_get_alloc_size(ggml_backend_buffer_type_t buft, const ggml_tensor* tensor) {
+    return ggml_backend_amx_get_alloc_size(tensor);
+
+    GGML_UNUSED(buft);
+}
+
+static bool ggml_backend_amx_buffer_type_is_host(ggml_backend_buffer_type_t buft) {
+    return false;
+
+    GGML_UNUSED(buft);
+}
+
+ggml_backend_buffer_type_t ggml_backend_amx_buffer_type() {
+    static struct ggml_backend_buffer_type ggml_backend_buffer_type_amx = {
+        /* .iface = */ {
+        /* .get_name         = */ ggml_backend_amx_buffer_type_get_name,
+        /* .alloc_buffer     = */ ggml_backend_amx_buffer_type_alloc_buffer,
+        /* .get_alignment    = */ ggml_backend_amx_buffer_type_get_alignment,
+        /* .get_max_size     = */ NULL, // defaults to SIZE_MAX
+        /* .get_alloc_size   = */ ggml_backend_amx_buffer_type_get_alloc_size,
+        /* .is_host          = */ ggml_backend_amx_buffer_type_is_host,
+        },
+        /* .device  = */ NULL,
+        /* .context = */ NULL,
+    };
+
+    return &ggml_backend_buffer_type_amx;
+}
+
+// backend interface
+
+static const char * ggml_backend_amx_name(ggml_backend_t backend) {
+    return "AMX";
+
+    GGML_UNUSED(backend);
+}
+
+static void ggml_backend_amx_free(ggml_backend_t backend) {
+    ggml_backend_amx_context * ctx = (ggml_backend_amx_context *)backend->context;
+    delete ctx;
+    delete backend;
+}
+
+static ggml_backend_buffer_type_t ggml_backend_amx_get_default_buffer_type(ggml_backend_t backend) {
+    return ggml_backend_amx_buffer_type();
+
+    GGML_UNUSED(backend);
+}
+
+static enum ggml_status ggml_backend_amx_graph_compute(ggml_backend_t backend, struct ggml_cgraph * cgraph) {
+    ggml_backend_amx_context * ctx = (ggml_backend_amx_context *)backend->context;
+
+    for (int i = 0; i < cgraph->n_nodes; i++) {
+        struct ggml_tensor * node = cgraph->nodes[i];
+
+        switch (node->op) {
+        case GGML_OP_MUL_MAT:
+            ggml_backend_amx_mul_mat(ctx, node);
+            break;
+
+        case GGML_OP_NONE:
+        case GGML_OP_RESHAPE:
+        case GGML_OP_VIEW:
+        case GGML_OP_PERMUTE:
+        case GGML_OP_TRANSPOSE:
+            break;
+
+        default:
+            fprintf(stderr, "%s: unsupported op %s\n", __func__, ggml_op_desc(node));
+            GGML_ASSERT(false);
+        }
+    }
+
+    return GGML_STATUS_SUCCESS;
+
+    GGML_UNUSED(backend);
+}
+
+static struct ggml_backend_i ggml_backend_amx_i = {
+    /* .get_name                = */ ggml_backend_amx_name,
+    /* .free                    = */ ggml_backend_amx_free,
+    /* .get_default_buffer_type = */ ggml_backend_amx_get_default_buffer_type,
+    /* .set_tensor_async        = */ NULL,
+    /* .get_tensor_async        = */ NULL,
+    /* .cpy_tensor_async        = */ NULL,
+    /* .synchronize             = */ NULL,
+    /* .graph_plan_create       = */ NULL,
+    /* .graph_plan_free         = */ NULL,
+    /* .graph_plan_update       = */ NULL,
+    /* .graph_plan_compute      = */ NULL,
+    /* .graph_compute           = */ ggml_backend_amx_graph_compute,
+    /* .supports_op             = */ NULL,
+    /* .supports_buft           = */ NULL,
+    /* .offload_op              = */ NULL,
+    /* .event_record            = */ NULL,
+    /* .event_wait              = */ NULL,
+};
+
+static ggml_guid_t ggml_backend_amx_guid() {
+    static ggml_guid guid = { 0x13, 0xb8, 0xa4, 0xc4, 0xba, 0xfe, 0x51, 0x67, 0x87, 0x44, 0x55, 0x15, 0xb2, 0x35, 0x62, 0x3e };
+    return &guid;
+}
+
+#define ARCH_GET_XCOMP_PERM     0x1022
+#define ARCH_REQ_XCOMP_PERM     0x1023
+#define XFEATURE_XTILECFG       17
+#define XFEATURE_XTILEDATA      18
+
+static bool ggml_amx_init() {
+#if defined(__gnu_linux__)
+    if (syscall(SYS_arch_prctl, ARCH_REQ_XCOMP_PERM, XFEATURE_XTILEDATA)) {
+        fprintf(stderr, "AMX is not ready to be used!\n");
+        return false;
+    }
+    return true;
+#elif defined(_WIN32)
+    return true;
+#endif
+}
+
+ggml_backend_t ggml_backend_amx_init() {
+
+    // invoke a Linux system call to request access to AMX features
+    ggml_amx_init();
+
+    // backend context
+    ggml_backend_amx_context * ctx = new ggml_backend_amx_context;
+
+    // ggml amx backend
+    ggml_backend_t backend = new ggml_backend {
+        /* .guid      = */ ggml_backend_amx_guid(),
+        /* .interface = */ ggml_backend_amx_i,
+        /* .device    = */ ggml_backend_reg_dev_get(ggml_backend_amx_reg(), 0),
+        /* .context   = */ ctx,
+    };
+
+    return backend;
+}
+
+bool ggml_backend_is_amx(ggml_backend_t backend) {
+    return backend != NULL && ggml_guid_matches(backend->guid, ggml_backend_amx_guid());
+}
+
+void ggml_backend_amx_set_n_threads(ggml_backend_t backend_amx, int n_threads) {
+    GGML_ASSERT(ggml_backend_is_amx(backend_amx));
+
+    ggml_backend_amx_context * ctx = (ggml_backend_amx_context *)backend_amx->context;
+    ctx->n_threads = n_threads;
+}
+
+// device interface
+
+static const char * ggml_backend_amx_device_get_name(ggml_backend_dev_t dev) {
+    return "AMX";
+
+    GGML_UNUSED(dev);
+}
+
+static const char * ggml_backend_amx_device_get_description(ggml_backend_dev_t dev) {
+    return "Intel Advanced Matrix Extensions";
+
+    GGML_UNUSED(dev);
+}
+
+static void ggml_backend_amx_device_get_memory(ggml_backend_dev_t dev, size_t * free, size_t * total) {
+    // TODO
+    *free = 0;
+    *total = 0;
+
+    GGML_UNUSED(dev);
+}
+
+static enum ggml_backend_dev_type ggml_backend_amx_device_get_type(ggml_backend_dev_t dev) {
+    return GGML_BACKEND_DEVICE_TYPE_CPU;
+
+    GGML_UNUSED(dev);
+}
+
+static void ggml_backend_amx_device_get_props(ggml_backend_dev_t dev, struct ggml_backend_dev_props * props) {
+    props->name        = ggml_backend_amx_device_get_name(dev);
+    props->description = ggml_backend_amx_device_get_description(dev);
+    props->type        = ggml_backend_amx_device_get_type(dev);
+    ggml_backend_amx_device_get_memory(dev, &props->memory_free, &props->memory_total);
+
+    // `buffer_from_host_ptr` is intended to be used in mmap, when memory layout unchanged
+    props->caps = {
+        /* .async                 = */ false,
+        /* .host_buffer           = */ false,
+        /* .buffer_from_host_ptr  = */ false,
+        /* .events                = */ false,
+    };
+}
+
+static ggml_backend_t ggml_backend_amx_device_init(ggml_backend_dev_t dev, const char * params) {
+    return ggml_backend_amx_init();
+
+    GGML_UNUSED(dev);
+    GGML_UNUSED(params);
+}
+
+static ggml_backend_buffer_type_t ggml_backend_amx_device_get_buffer_type(ggml_backend_dev_t dev) {
+    return ggml_backend_amx_buffer_type();
+
+    GGML_UNUSED(dev);
+}
+
+static bool ggml_backend_amx_device_supports_op(ggml_backend_dev_t dev, const struct ggml_tensor * op) {
+
+    // handle only 2d gemm for now
+    auto is_contiguous_2d = [](const struct ggml_tensor * t) {
+        return ggml_is_contiguous(t) && t->ne[3] == 1 && t->ne[2] == 1;
+    };
+
+    switch (op->op) {
+        case GGML_OP_NONE:
+        case GGML_OP_RESHAPE:
+        case GGML_OP_VIEW:
+        case GGML_OP_PERMUTE:
+        case GGML_OP_TRANSPOSE:
+            return true;
+
+        case GGML_OP_MUL_MAT: {
+            const struct ggml_tensor * src0 = op->src[0];
+            const struct ggml_tensor * src1 = op->src[1];
+
+            const enum ggml_type type = src0->type;
+            const int64_t ne0 = op->ne[0];
+
+            bool is_training = src0->grad || src1->grad;
+
+            // amx kernels enables for Q4_0, Q4_1, Q8_0, F16
+            // Q4_K, Q5_K, Q6_K, IQ4_XS enabled for QK_K = 256
+            bool has_amx_kernels = qtype_has_amx_kernels(type) || (type == GGML_TYPE_F16);
+
+            bool can_use_amx =
+                is_contiguous_2d(src0) &&       // src0 must be contiguous
+                is_contiguous_2d(src1) &&       // src1 must be contiguous
+                !is_training &&                 // inference only
+                src1->type == GGML_TYPE_F32 &&  // src1 must be float32
+                has_amx_kernels &&              // with amx kernel impls
+                ne0 % (TILE_N * 2) == 0;        // out_features is 32x
+
+            return can_use_amx;
+        }
+        default:
+            return false;
+    }
+
+    GGML_UNUSED(dev);
+}
+
+static bool ggml_backend_amx_device_supports_buft(ggml_backend_dev_t dev, ggml_backend_buffer_type_t buft) {
+    return buft->iface.get_name == ggml_backend_amx_buffer_type_get_name;
+
+    GGML_UNUSED(dev);
+}
+
+static const struct ggml_backend_device_i ggml_backend_amx_device_i = {
+    /* .get_name             = */ ggml_backend_amx_device_get_name,
+    /* .get_description      = */ ggml_backend_amx_device_get_description,
+    /* .get_memory           = */ ggml_backend_amx_device_get_memory,
+    /* .get_type             = */ ggml_backend_amx_device_get_type,
+    /* .get_props            = */ ggml_backend_amx_device_get_props,
+    /* .init_backend         = */ ggml_backend_amx_device_init,
+    /* .get_buffer_type      = */ ggml_backend_amx_device_get_buffer_type,
+    /* .get_host_buffer_type = */ NULL,
+    /* .buffer_from_host_ptr = */ NULL,
+    /* .supports_op          = */ ggml_backend_amx_device_supports_op,
+    /* .supports_buft        = */ ggml_backend_amx_device_supports_buft,
+    /* .offload_op           = */ NULL,
+    /* .event_new            = */ NULL,
+    /* .event_free           = */ NULL,
+    /* .event_synchronize    = */ NULL,
+};
+
+// backend reg interface
+
+static const char * ggml_backend_amx_reg_get_name(ggml_backend_reg_t reg) {
+    return "AMX";
+
+    GGML_UNUSED(reg);
+}
+
+static size_t ggml_backend_amx_reg_get_device_count(ggml_backend_reg_t reg) {
+    return 1;
+
+    GGML_UNUSED(reg);
+}
+
+static ggml_backend_dev_t ggml_backend_amx_reg_get_device(ggml_backend_reg_t reg, size_t index) {
+    GGML_ASSERT(index == 0);
+
+    static ggml_backend_device ggml_backend_amx_device = {
+        /* .iface   = */ ggml_backend_amx_device_i,
+        /* .reg     = */ reg,
+        /* .context = */ nullptr,
+    };
+
+    return &ggml_backend_amx_device;
+
+    GGML_UNUSED(reg);
+    GGML_UNUSED(index);
+}
+
+static void * ggml_backend_amx_get_proc_address(ggml_backend_reg_t reg, const char * name) {
+    if (std::strcmp(name, "ggml_backend_set_n_threads") == 0) {
+        return (void *)ggml_backend_amx_set_n_threads;
+    }
+    return NULL;
+
+    GGML_UNUSED(reg);
+    GGML_UNUSED(name);
+}
+
+static const struct ggml_backend_reg_i ggml_backend_amx_reg_i = {
+    /* .get_name         = */ ggml_backend_amx_reg_get_name,
+    /* .get_device_count = */ ggml_backend_amx_reg_get_device_count,
+    /* .get_device       = */ ggml_backend_amx_reg_get_device,
+    /* .get_proc_address = */ ggml_backend_amx_get_proc_address,
+};
+
+ggml_backend_reg_t ggml_backend_amx_reg(void) {
+    static struct ggml_backend_reg ggml_backend_amx_reg = {
+        /* .iface   = */ ggml_backend_amx_reg_i,
+        /* .context = */ NULL,
+    };
+
+    return &ggml_backend_amx_reg;
+}
+
+#else // if defined(__AMX_INT8__)
+
+ggml_backend_t ggml_backend_amx_init(void) {
+    fprintf(stderr, "GGML is not compiled with AMX support!\n");
+    return ggml_backend_t{};
+}
+
+void ggml_backend_amx_set_n_threads(ggml_backend_t backend_amx, int n_threads) {
+    fprintf(stderr, "GGML is not compiled with AMX support!\n");
+
+    GGML_UNUSED(backend_amx);
+    GGML_UNUSED(n_threads);
+}
+
+#endif
diff --git a/ggml/src/ggml-amx/common.h b/ggml/src/ggml-amx/common.h
new file mode 100644
index 000000000..2b6c63527
--- /dev/null
+++ b/ggml/src/ggml-amx/common.h
@@ -0,0 +1,93 @@
+#pragma once
+
+#include "ggml.h"
+#include "ggml-cpu-impl.h" // 
+
+#include 
+#include 
+#include 
+
+#if defined(_OPENMP)
+#include 
+#endif
+
+#define TILE_M 16
+#define TILE_N 16
+#define TILE_K 32
+#define VNNI_BLK 4
+
+#define AMX_BLK_SIZE 32
+
+#define TMM0 0
+#define TMM1 1
+#define TMM2 2
+#define TMM3 3
+#define TMM4 4
+#define TMM5 5
+#define TMM6 6
+#define TMM7 7
+
+// parallel routines
+template ::value, int>::type = 0>
+inline T div_up(T x, T y) { return (x + y - 1) / y; }
+
+template 
+inline void balance211(T n, T nth, T ith, T& n_start, T& n_end) {
+#if 0
+    // onednn partition pattern
+    T& n_my = n_end;
+    if (nth <= 1 || n == 0) {
+        n_start = 0;
+        n_my = n;
+    } else {
+        T n1 = div_up(n, nth);
+        T n2 = n1 - 1;
+        T T1 = n - n2 * nth;
+        n_my = ith < T1 ? n1 : n2;
+        n_start = ith <= T1 ? ith*n1 : T1 * n1 + (ith - T1) * n2;
+    }
+    n_end += n_start;
+#else
+    // pytorch aten partition pattern
+    T n_my = div_up(n, nth);
+    n_start = ith * n_my;
+    n_end = std::min(n_start + n_my, n);
+#endif
+}
+
+template 
+inline void parallel_for(int nth, int n, const func_t& f) {
+#if defined(_OPENMP)
+#pragma omp parallel num_threads(nth)
+{
+    //int nth = omp_get_num_threads();
+    int ith = omp_get_thread_num();
+    int tbegin, tend;
+    balance211(n, nth, ith, tbegin, tend);
+    f(tbegin, tend);
+}
+#else
+    f(0, n);
+
+    GGML_UNUSED(nth);
+#endif
+}
+
+// quantized types that have AMX support
+inline bool qtype_has_amx_kernels(const enum ggml_type type) {
+    // TODO: fix padding for vnni format
+    return (type == GGML_TYPE_Q4_0) ||
+        (type == GGML_TYPE_Q4_1);
+        //(type == GGML_TYPE_Q8_0) ||
+        //(type == GGML_TYPE_Q4_K) ||
+        //(type == GGML_TYPE_Q5_K) ||
+        //(type == GGML_TYPE_Q6_K) ||
+        //(type == GGML_TYPE_IQ4_XS);
+}
+
+// ggml backend context
+struct ggml_backend_amx_context {
+    int n_threads = GGML_DEFAULT_N_THREADS;
+    std::unique_ptr work_data;
+    size_t work_size = 0;
+};
diff --git a/ggml/src/ggml-amx/mmq.cpp b/ggml/src/ggml-amx/mmq.cpp
new file mode 100644
index 000000000..239d15121
--- /dev/null
+++ b/ggml/src/ggml-amx/mmq.cpp
@@ -0,0 +1,2509 @@
+
+#if defined(__GNUC__)
+#pragma GCC diagnostic ignored "-Wpedantic"
+#pragma GCC diagnostic ignored "-Wunused-local-typedefs"
+#endif
+
+#include "mmq.h"
+#include "ggml-impl.h"
+#include "ggml-quants.h"
+#include 
+#include 
+
+#if defined(__gnu_linux__)
+#include 
+#include 
+#endif
+
+#if defined(_OPENMP)
+#include 
+#endif
+
+#if (defined(_WIN32) || defined(_WIN64))
+#define RESTRICT __restrict
+#else
+#define RESTRICT __restrict__
+#endif
+
+#if (defined(_WIN32) || defined(_WIN64))
+#define ALWAYS_INLINE __forceinline
+#elif __has_attribute(always_inline) || defined(__GNUC__)
+#define ALWAYS_INLINE __attribute__((__always_inline__)) inline
+#else
+#define ALWAYS_INLINE inline
+#endif
+
+#if defined(__AMX_INT8__)
+
+namespace {
+
+// Forced unrolling
+template 
+struct Unroll {
+    template 
+    ALWAYS_INLINE void operator()(const Func& f, Args... args) const {
+        Unroll{}(f, args...);
+        f(std::integral_constant{}, args...);
+    }
+};
+
+template <>
+struct Unroll<1> {
+    template 
+    ALWAYS_INLINE void operator()(const Func& f, Args... args) const {
+        f(std::integral_constant{}, args...);
+    }
+};
+
+// type traits
+template  struct PackedTypes {};
+template <> struct PackedTypes { using type = int8_t; };
+template <> struct PackedTypes { using type = uint8_t; };
+template <> struct PackedTypes { using type = int8_t; };
+template  using packed_B_type = typename PackedTypes::type;
+
+template 
+struct do_compensate : std::integral_constant::value> {};
+
+template 
+struct do_unpack : std::integral_constant::value ||
+    std::is_same::value> {};
+
+template 
+struct is_type_qkk : std::integral_constant::value ||
+    std::is_same::value ||
+    std::is_same::value ||
+    std::is_same::value> {};
+
+#define GGML_DISPATCH_FLOATING_TYPES(TYPE, ...)                                        \
+    [&] {                                                                              \
+        switch (TYPE) {                                                                \
+            case GGML_TYPE_F16: {                                                      \
+                using type = ggml_fp16_t;                                              \
+                constexpr int blck_size = 16;                                          \
+                return __VA_ARGS__();                                                  \
+            }                                                                          \
+            case GGML_TYPE_BF16: {                                                     \
+                using type = ggml_bf16_t;                                              \
+                constexpr int blck_size = 32;                                          \
+                return __VA_ARGS__();                                                  \
+            }                                                                          \
+            default:                                                                   \
+                fprintf(stderr, "Unsupported floating data type\n");                   \
+        }                                                                              \
+    }()
+
+#define GGML_DISPATCH_QTYPES(QT, ...)                                                  \
+    [&] {                                                                              \
+        switch (QT) {                                                                  \
+            case GGML_TYPE_Q4_0: {                                                     \
+                using type = block_q4_0;                                               \
+                using vec_dot_type = block_q8_0;                                       \
+                constexpr int blck_size = QK4_0;                                       \
+                return __VA_ARGS__();                                                  \
+            }                                                                          \
+            case GGML_TYPE_Q4_1: {                                                     \
+                using type = block_q4_1;                                               \
+                using vec_dot_type = block_q8_1;                                       \
+                constexpr int blck_size = QK4_1;                                       \
+                return __VA_ARGS__();                                                  \
+            }                                                                          \
+            case GGML_TYPE_Q8_0: {                                                     \
+                using type = block_q8_0;                                               \
+                using vec_dot_type = block_q8_0;                                       \
+                constexpr int blck_size = QK8_0;                                       \
+                return __VA_ARGS__();                                                  \
+            }                                                                          \
+            case GGML_TYPE_Q4_K: {                                                     \
+                using type = block_q4_K;                                               \
+                using vec_dot_type = block_q8_K;                                       \
+                constexpr int blck_size = QK_K;                                        \
+                return __VA_ARGS__();                                                  \
+            }                                                                          \
+            case GGML_TYPE_Q5_K: {                                                     \
+                using type = block_q5_K;                                               \
+                using vec_dot_type = block_q8_K;                                       \
+                constexpr int blck_size = QK_K;                                        \
+                return __VA_ARGS__();                                                  \
+            }                                                                          \
+            case GGML_TYPE_Q6_K: {                                                     \
+                using type = block_q6_K;                                               \
+                using vec_dot_type = block_q8_K;                                       \
+                constexpr int blck_size = QK_K;                                        \
+                return __VA_ARGS__();                                                  \
+            }                                                                          \
+            case GGML_TYPE_IQ4_XS: {                                                   \
+                using type = block_iq4_xs;                                             \
+                using vec_dot_type = block_q8_K;                                       \
+                constexpr int blck_size = QK_K;                                        \
+                return __VA_ARGS__();                                                  \
+            }                                                                          \
+            default:                                                                   \
+                fprintf(stderr, "Unsupported quantized data type: %d\n", int(TYPE));   \
+        }                                                                              \
+    }()
+
+#define GGML_DISPATCH_BOOL(BOOL_V, BOOL_NAME, ...)                                     \
+    [&] {                                                                              \
+        if (BOOL_V) {                                                                  \
+            constexpr bool BOOL_NAME = true;                                           \
+            return __VA_ARGS__();                                                      \
+        } else {                                                                       \
+            constexpr bool BOOL_NAME = false;                                          \
+            return __VA_ARGS__();                                                      \
+        }                                                                              \
+    }()
+
+// define amx tile config data structure
+struct tile_config_t{
+    uint8_t palette_id = 0;
+    uint8_t start_row = 0;
+    uint8_t reserved_0[14] = {0};
+    uint16_t colsb[16] = {0};
+    uint8_t rows[16] = {0};
+};
+
+// Notes: amx tile config
+//
+// Typically, TMUL calculates A and B of size 16 x 64 containing INT8 values,
+// and accumulate the result to a 16 x 16 matrix C containing INT32 values,
+//
+// As many GGUF quantized types as `block_size` of 32, so a 16-16-32 config is used
+// instead of the normally used 16-16-64 config.
+//
+//    Block A: {16, 32}, dtype = int8_t
+//    Block B: {16, 32}, dtype = uint8_t/int8_t
+//    Block C: {16, 16}, dtype = int32_t
+//
+// Block B needs to be prepacked to vnni format before feeding into  TMUL:
+//    packed_B: from {n, k} to {k/vnni_blk, n, vnni_blck}, viewed in 2d, we get {8, 64}
+//
+// Therefore, we get tileconfig:
+//             A    B    C
+//    rows    16    8   16
+//    colsb   32   64   16
+//
+// For tile distribution, follow a 2-2-4 pattern, e.g. A used TMM2-TMM3, B used TMM0-TMM1,
+// C used TMM4-TMM7:
+//            B TMM0  B TMM1
+//    A TMM2  C TMM4  C TMM6
+//    A TMM3  C TMM5  C TMM7
+//
+// Each `amx` kernel handles 4 blocks at a time: 2MB * 2NB, when m < 2 * BLOCK_M, unpack A
+// will be needed.
+//
+// Here another commonly used pattern 1-3-3 is skipped, as it is mostly used when m <=16;
+// and the sinlge batch gemm (m=1) has a special fast path with `avx512-vnni`.
+//
+// ref: https://www.intel.com/content/www/us/en/developer/articles/code-sample/
+//    advanced-matrix-extensions-intrinsics-functions.html
+//
+
+#define TC_CONFIG_TILE(i, r, cb) tc.rows[i] = r; tc.colsb[i] = cb
+void ggml_tile_config_init(void) {
+    static thread_local bool is_first_time = true;
+
+    if (!is_first_time) {
+        return;
+    }
+
+    static thread_local tile_config_t tc;
+    tile_config_t current_tc;
+    _tile_storeconfig(¤t_tc);
+
+    // load only when config changes
+    if (tc.palette_id == 0 || (memcmp(¤t_tc.colsb, &tc.colsb, sizeof(uint16_t) * 8) != 0 &&
+                               memcmp(¤t_tc.rows, &tc.rows, sizeof(uint8_t) * 8) != 0)) {
+        tc.palette_id = 1;
+        tc.start_row = 0;
+        TC_CONFIG_TILE(TMM0, 8, 64);
+        TC_CONFIG_TILE(TMM1, 8, 64);
+        TC_CONFIG_TILE(TMM2, 16, 32);
+        TC_CONFIG_TILE(TMM3, 16, 32);
+        TC_CONFIG_TILE(TMM4, 16, 64);
+        TC_CONFIG_TILE(TMM5, 16, 64);
+        TC_CONFIG_TILE(TMM6, 16, 64);
+        TC_CONFIG_TILE(TMM7, 16, 64);
+        _tile_loadconfig(&tc);
+    }
+
+    is_first_time = false;
+}
+
+// we need an extra 16 * 4B (TILE_N * int32_t) for each NB/KB block for compensation.
+// See the notes `s8s8 igemm compensation in avx512-vnni` for detail.
+template 
+int get_tile_size() {
+    int tile_size = TILE_N * sizeof(TB);
+    if (do_compensate::value) {
+        tile_size += TILE_N * sizeof(int32_t);
+    }
+    if (std::is_same::value ||
+        std::is_same::value) {
+        tile_size += TILE_N * 4;
+    }
+    if (std::is_same::value) {
+        tile_size += TILE_N * 2;
+    }
+    return tile_size;
+}
+
+template 
+int get_row_size(int K) {
+    int KB = K / BLOCK_K;
+    int row_size = KB * sizeof(TB);
+    if (do_compensate::value) {
+        row_size += KB * sizeof(int32_t);
+    }
+    if (std::is_same::value ||
+        std::is_same::value) {
+        row_size += KB * 4;
+    }
+    if (std::is_same::value) {
+        row_size += KB * 2;
+    }
+    return row_size;
+}
+
+// vectorized dtype conversion
+inline float FP16_TO_FP32(ggml_half val) {
+    __m256i v = _mm256_setr_epi16(
+        val, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
+    __m512 o = _mm512_cvtph_ps(v);
+    return _mm512_cvtss_f32(o);
+}
+
+inline __m512 FP16_TO_FP32_VEC(ggml_half val) {
+    __m256i v = _mm256_set1_epi16(val);
+    return _mm512_cvtph_ps(v);
+}
+
+// horizontal reduce
+inline float _mm512_reduce_max_ps(const __m512 x) {
+    __m512 v = x;
+    __m512 v1 = _mm512_shuffle_f32x4(v, v, 0x4E);
+    v = _mm512_max_ps(v, v1);
+    v1 = _mm512_shuffle_f32x4(v, v, 0xB1);
+    v = _mm512_max_ps(v, v1);
+    v1 = _mm512_shuffle_ps(v, v, 0x4E);
+    v = _mm512_max_ps(v, v1);
+    v1 = _mm512_shuffle_ps(v, v, 0xB1);
+    v = _mm512_max_ps(v, v1);
+    return _mm512_cvtss_f32(v);
+}
+
+// transpose utils
+#define SHUFFLE_EPI32(a, b, mask) \
+    _mm256_castps_si256(_mm256_shuffle_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b), mask))
+inline void transpose_8x8_32bit(__m256i * v, __m256i * v1) {
+    // unpacking and 32-bit elements
+    v1[0] = _mm256_unpacklo_epi32(v[0], v[1]);
+    v1[1] = _mm256_unpackhi_epi32(v[0], v[1]);
+    v1[2] = _mm256_unpacklo_epi32(v[2], v[3]);
+    v1[3] = _mm256_unpackhi_epi32(v[2], v[3]);
+    v1[4] = _mm256_unpacklo_epi32(v[4], v[5]);
+    v1[5] = _mm256_unpackhi_epi32(v[4], v[5]);
+    v1[6] = _mm256_unpacklo_epi32(v[6], v[7]);
+    v1[7] = _mm256_unpackhi_epi32(v[6], v[7]);
+
+    // shuffling the 32-bit elements
+    v[0] = SHUFFLE_EPI32(v1[0], v1[2], 0x44);
+    v[1] = SHUFFLE_EPI32(v1[0], v1[2], 0xee);
+    v[2] = SHUFFLE_EPI32(v1[4], v1[6], 0x44);
+    v[3] = SHUFFLE_EPI32(v1[4], v1[6], 0xee);
+    v[4] = SHUFFLE_EPI32(v1[1], v1[3], 0x44);
+    v[5] = SHUFFLE_EPI32(v1[1], v1[3], 0xee);
+    v[6] = SHUFFLE_EPI32(v1[5], v1[7], 0x44);
+    v[7] = SHUFFLE_EPI32(v1[5], v1[7], 0xee);
+
+    // shuffling 128-bit elements
+    v1[0] = _mm256_permute2f128_si256(v[2], v[0], 0x02);
+    v1[1] = _mm256_permute2f128_si256(v[3], v[1], 0x02);
+    v1[2] = _mm256_permute2f128_si256(v[6], v[4], 0x02);
+    v1[3] = _mm256_permute2f128_si256(v[7], v[5], 0x02);
+    v1[4] = _mm256_permute2f128_si256(v[2], v[0], 0x13);
+    v1[5] = _mm256_permute2f128_si256(v[3], v[1], 0x13);
+    v1[6] = _mm256_permute2f128_si256(v[6], v[4], 0x13);
+    v1[7] = _mm256_permute2f128_si256(v[7], v[5], 0x13);
+}
+
+inline void transpose_16x4_32bit(__m512i * r, __m512i * d) {
+
+    static const __m512i index1 = _mm512_set_epi32(
+        0x0f, 0x0b, 0x07, 0x03,
+        0x0e, 0x0a, 0x06, 0x02,
+        0x0d, 0x09, 0x05, 0x01,
+        0x0c, 0x08, 0x04, 0x00);
+
+    d[0] = _mm512_permutexvar_epi32(index1, r[0]);
+    d[1] = _mm512_permutexvar_epi32(index1, r[1]);
+    d[2] = _mm512_permutexvar_epi32(index1, r[2]);
+    d[3] = _mm512_permutexvar_epi32(index1, r[3]);
+
+    r[0] = _mm512_shuffle_i32x4(d[0], d[1], 0x44);
+    r[1] = _mm512_shuffle_i32x4(d[0], d[1], 0xee);
+    r[2] = _mm512_shuffle_i32x4(d[2], d[3], 0x44);
+    r[3] = _mm512_shuffle_i32x4(d[2], d[3], 0xee);
+
+    d[0] = _mm512_shuffle_i32x4(r[0], r[2], 0x88);
+    d[1] = _mm512_shuffle_i32x4(r[0], r[2], 0xdd);
+    d[2] = _mm512_shuffle_i32x4(r[1], r[3], 0x88);
+    d[3] = _mm512_shuffle_i32x4(r[1], r[3], 0xdd);
+}
+
+inline void transpose_16x16_32bit(__m512i * v) {
+    __m512i v1[16];
+    v1[0] = _mm512_unpacklo_epi32(v[0], v[1]);
+    v1[1] = _mm512_unpackhi_epi32(v[0], v[1]);
+    v1[2] = _mm512_unpacklo_epi32(v[2], v[3]);
+    v1[3] = _mm512_unpackhi_epi32(v[2], v[3]);
+    v1[4] = _mm512_unpacklo_epi32(v[4], v[5]);
+    v1[5] = _mm512_unpackhi_epi32(v[4], v[5]);
+    v1[6] = _mm512_unpacklo_epi32(v[6], v[7]);
+    v1[7] = _mm512_unpackhi_epi32(v[6], v[7]);
+    v1[8] = _mm512_unpacklo_epi32(v[8], v[9]);
+    v1[9] = _mm512_unpackhi_epi32(v[8], v[9]);
+    v1[10] = _mm512_unpacklo_epi32(v[10], v[11]);
+    v1[11] = _mm512_unpackhi_epi32(v[10], v[11]);
+    v1[12] = _mm512_unpacklo_epi32(v[12], v[13]);
+    v1[13] = _mm512_unpackhi_epi32(v[12], v[13]);
+    v1[14] = _mm512_unpacklo_epi32(v[14], v[15]);
+    v1[15] = _mm512_unpackhi_epi32(v[14], v[15]);
+
+    v[0] = _mm512_unpacklo_epi64(v1[0], v1[2]);
+    v[1] = _mm512_unpackhi_epi64(v1[0], v1[2]);
+    v[2] = _mm512_unpacklo_epi64(v1[1], v1[3]);
+    v[3] = _mm512_unpackhi_epi64(v1[1], v1[3]);
+    v[4] = _mm512_unpacklo_epi64(v1[4], v1[6]);
+    v[5] = _mm512_unpackhi_epi64(v1[4], v1[6]);
+    v[6] = _mm512_unpacklo_epi64(v1[5], v1[7]);
+    v[7] = _mm512_unpackhi_epi64(v1[5], v1[7]);
+    v[8] = _mm512_unpacklo_epi64(v1[8], v1[10]);
+    v[9] = _mm512_unpackhi_epi64(v1[8], v1[10]);
+    v[10] = _mm512_unpacklo_epi64(v1[9], v1[11]);
+    v[11] = _mm512_unpackhi_epi64(v1[9], v1[11]);
+    v[12] = _mm512_unpacklo_epi64(v1[12], v1[14]);
+    v[13] = _mm512_unpackhi_epi64(v1[12], v1[14]);
+    v[14] = _mm512_unpacklo_epi64(v1[13], v1[15]);
+    v[15] = _mm512_unpackhi_epi64(v1[13], v1[15]);
+
+    v1[0] = _mm512_shuffle_i32x4(v[0], v[4], 0x88);
+    v1[1] = _mm512_shuffle_i32x4(v[1], v[5], 0x88);
+    v1[2] = _mm512_shuffle_i32x4(v[2], v[6], 0x88);
+    v1[3] = _mm512_shuffle_i32x4(v[3], v[7], 0x88);
+    v1[4] = _mm512_shuffle_i32x4(v[0], v[4], 0xdd);
+    v1[5] = _mm512_shuffle_i32x4(v[1], v[5], 0xdd);
+    v1[6] = _mm512_shuffle_i32x4(v[2], v[6], 0xdd);
+    v1[7] = _mm512_shuffle_i32x4(v[3], v[7], 0xdd);
+    v1[8] = _mm512_shuffle_i32x4(v[8], v[12], 0x88);
+    v1[9] = _mm512_shuffle_i32x4(v[9], v[13], 0x88);
+    v1[10] = _mm512_shuffle_i32x4(v[10], v[14], 0x88);
+    v1[11] = _mm512_shuffle_i32x4(v[11], v[15], 0x88);
+    v1[12] = _mm512_shuffle_i32x4(v[8], v[12], 0xdd);
+    v1[13] = _mm512_shuffle_i32x4(v[9], v[13], 0xdd);
+    v1[14] = _mm512_shuffle_i32x4(v[10], v[14], 0xdd);
+    v1[15] = _mm512_shuffle_i32x4(v[11], v[15], 0xdd);
+
+    v[0] = _mm512_shuffle_i32x4(v1[0], v1[8], 0x88);
+    v[1] = _mm512_shuffle_i32x4(v1[1], v1[9], 0x88);
+    v[2] = _mm512_shuffle_i32x4(v1[2], v1[10], 0x88);
+    v[3] = _mm512_shuffle_i32x4(v1[3], v1[11], 0x88);
+    v[4] = _mm512_shuffle_i32x4(v1[4], v1[12], 0x88);
+    v[5] = _mm512_shuffle_i32x4(v1[5], v1[13], 0x88);
+    v[6] = _mm512_shuffle_i32x4(v1[6], v1[14], 0x88);
+    v[7] = _mm512_shuffle_i32x4(v1[7], v1[15], 0x88);
+    v[8] = _mm512_shuffle_i32x4(v1[0], v1[8], 0xdd);
+    v[9] = _mm512_shuffle_i32x4(v1[1], v1[9], 0xdd);
+    v[10] = _mm512_shuffle_i32x4(v1[2], v1[10], 0xdd);
+    v[11] = _mm512_shuffle_i32x4(v1[3], v1[11], 0xdd);
+    v[12] = _mm512_shuffle_i32x4(v1[4], v1[12], 0xdd);
+    v[13] = _mm512_shuffle_i32x4(v1[5], v1[13], 0xdd);
+    v[14] = _mm512_shuffle_i32x4(v1[6], v1[14], 0xdd);
+    v[15] = _mm512_shuffle_i32x4(v1[7], v1[15], 0xdd);
+}
+
+void quantize_row_q8_K_vnni(const float * RESTRICT x, void * RESTRICT vy, int64_t k) {
+    assert(k % QK_K == 0);
+    const int KB = k / QK_K;
+    constexpr int kVecs = QK_K / 16;
+
+    block_q8_K * y = reinterpret_cast(vy);
+
+    // hold 16 float vecs from x
+    __m512  v[kVecs];
+
+    // hold the quants vecs
+    __m512i vq[kVecs / 4];
+
+    // hold the packed quants vecs
+    __m512i vq_packed[kVecs / 4];
+
+    const __m512 signBit = _mm512_set1_ps(-0.f);
+
+    for (int i = 0; i < KB; ++i) {
+        // Compute max(abs(e)) for the block
+        __m512 vamax = _mm512_set1_ps(0.f);
+        for (int j = 0; j < kVecs; ++j) {
+            v[j] = _mm512_loadu_ps(x); x += 16;
+            vamax = _mm512_max_ps(vamax, _mm512_andnot_ps(signBit, v[j]));
+        }
+        const float amax = _mm512_reduce_max_ps(vamax);
+
+        // Quantize these floats
+        const float iscale = 127.f / amax;
+        y[i].d = GGML_FP32_TO_FP16(1 / iscale);
+        const float id = ( amax != 0.0f ) ? iscale : 0.f;
+        const __m512 vscale = _mm512_set1_ps(id);
+
+        // Apply multiplier and round to nearest integer
+        for (int j = 0; j < kVecs; ++j) {
+            v[j] = _mm512_mul_ps(v[j], vscale);
+            v[j] = _mm512_roundscale_ps(v[j], (_MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC));
+        }
+
+        // Pack to epi8 vecs
+        for (int j = 0; j < kVecs / 4; ++j) {
+            __m128i q8_0 = _mm512_cvtepi32_epi8(_mm512_cvtps_epi32(v[j * 4 + 0]));
+            __m128i q8_1 = _mm512_cvtepi32_epi8(_mm512_cvtps_epi32(v[j * 4 + 1]));
+            __m128i q8_2 = _mm512_cvtepi32_epi8(_mm512_cvtps_epi32(v[j * 4 + 2]));
+            __m128i q8_3 = _mm512_cvtepi32_epi8(_mm512_cvtps_epi32(v[j * 4 + 3]));
+
+            __m256i q8_01 = _mm256_insertf128_si256(_mm256_castsi128_si256(q8_0), (q8_1), 1);
+            __m256i q8_23 = _mm256_insertf128_si256(_mm256_castsi128_si256(q8_2), (q8_3), 1);
+
+            vq[j] = _mm512_inserti32x8(_mm512_castsi256_si512(q8_01), q8_23, 1);
+            _mm512_storeu_si512((__m512i *)(y[i].qs + j * 64), vq[j]);
+        }
+
+        // Compute the bsums with vnni
+        transpose_16x4_32bit(vq, vq_packed);
+
+        const __m512i one = _mm512_set1_epi8(1);
+        __m512i sum = _mm512_setzero_si512();
+        for (int k = 0; k < 4; ++k) {
+            sum = _mm512_dpbusd_epi32(sum, one, vq_packed[k]);
+        }
+        _mm256_storeu_si256((__m256i *)(y[i].bsums), _mm512_cvtepi32_epi16(sum));
+    }
+}
+
+// quantize A from float to `vec_dot_type`
+template 
+inline void from_float(const float * x, char * vy, int64_t k);
+
+template <>
+inline void from_float(const float * x, char * vy, int64_t k) {
+    quantize_row_q8_0(x, vy, k);
+}
+
+template <>
+inline void from_float(const float * x, char * vy, int64_t k) {
+    quantize_row_q8_1(x, vy, k);
+}
+
+template <>
+inline void from_float(const float * x, char * vy, int64_t k) {
+#if 1
+    // TODO: this is reference impl!
+    quantize_row_q8_K(x, vy, k);
+#else
+    quantize_row_q8_K_vnni(x, vy, k);
+#endif
+}
+
+// load A from memory to array when nrows can not fill in whole tile
+void unpack_A(int8_t * RESTRICT tile, const block_q8_0 * RESTRICT A, int lda, int nr) {
+    assert(nr != TILE_M);
+    for (int m = 0; m < nr; ++m) {
+        const __m256i v = _mm256_loadu_si256((const __m256i *)(A[m * lda].qs));
+        _mm256_storeu_si256((__m256i *)(tile + m * TILE_K), v);
+    }
+}
+
+void unpack_A(int8_t * RESTRICT tile, const block_q8_1 * RESTRICT A, int lda, int nr) {
+    assert(nr != TILE_M);
+    for (int m = 0; m < nr; ++m) {
+        const __m256i v = _mm256_loadu_si256((const __m256i *)(A[m * lda].qs));
+        _mm256_storeu_si256((__m256i *)(tile + m * TILE_K), v);
+    }
+}
+
+template 
+void unpack_A(int8_t * RESTRICT tile, const block_q8_K * RESTRICT A, int lda, int k, int nr) {
+    assert(nr <= TILE_M);
+    for (int m = 0; m < nr; ++m) {
+        const __m256i v = _mm256_loadu_si256((const __m256i *)(A[m * lda].qs + k * 32));
+        _mm256_storeu_si256((__m256i *)(tile + m * TILE_K), v);
+    }
+}
+
+template <>
+void unpack_A(int8_t * RESTRICT tile, const block_q8_K * RESTRICT A, int lda, int k, int nr) {
+    assert(nr <= TILE_M);
+    // zero padding k from 16 to 32, so that we don't have to re-config amx
+    const __m128i zero = _mm_setzero_si128();
+    for (int m = 0; m < nr; ++m) {
+        const __m128i v = _mm_loadu_si128((const __m128i *)(A[m * lda].qs + k * 16));
+        const __m256i r = _mm256_insertf128_si256(_mm256_castsi128_si256(v), zero, 1);
+        _mm256_storeu_si256((__m256i *)(tile + m * TILE_K), r);
+    }
+}
+
+#define MM256_SET_M128I(a, b) _mm256_insertf128_si256(_mm256_castsi128_si256(b), (a), 1)
+inline __m256i bytes_from_nibbles_32(const uint8_t * rsi) {
+    const __m128i tmp = _mm_loadu_si128((const __m128i *)rsi);
+    const __m256i bytes = MM256_SET_M128I(_mm_srli_epi16(tmp, 4), tmp);
+    const __m256i lowMask = _mm256_set1_epi8(0xF);
+    return _mm256_and_si256(lowMask, bytes);
+}
+
+// used for block_q4_K
+inline __m512i bytes_from_nibbles_64(const uint8_t * rsi) {
+    const __m256i tmp = _mm256_loadu_si256((const __m256i *)rsi);
+    const __m256i lowMask = _mm256_set1_epi8(0xF);
+    const __m256i q4l = _mm256_and_si256(tmp, lowMask);
+    const __m256i q4h = _mm256_and_si256(_mm256_srli_epi16(tmp, 4), lowMask);
+    return _mm512_inserti32x8(_mm512_castsi256_si512(q4l), q4h, 1);
+}
+
+// used for block_q5_K
+inline __m512i bytes_from_nibbles_64(const uint8_t * qs, const uint8_t * qh, int k) {
+    const __m256i lowMask = _mm256_set1_epi8(0xF);
+    __m256i hmask = _mm256_set1_epi8(1);
+    hmask = _mm256_slli_epi16(hmask, k);
+
+    const __m256i q5bits = _mm256_loadu_si256((const __m256i *)qs);
+    const __m256i hbits = _mm256_loadu_si256((const __m256i *)qh);
+
+    const __m256i q5l_0 = _mm256_and_si256(q5bits, lowMask);
+    const __m256i q5h_0 = _mm256_slli_epi16(_mm256_srli_epi16(_mm256_and_si256(hbits, hmask), k + 0), 4);
+    const __m256i q5_0  = _mm256_add_epi8(q5l_0, q5h_0);
+    hmask = _mm256_slli_epi16(hmask, 1);
+
+    const __m256i q5l_1 = _mm256_and_si256(_mm256_srli_epi16(q5bits, 4), lowMask);
+    const __m256i q5h_1 = _mm256_slli_epi16(_mm256_srli_epi16(_mm256_and_si256(hbits, hmask), k + 1), 4);
+    const __m256i q5_1  = _mm256_add_epi8(q5l_1, q5h_1);
+
+    return _mm512_inserti32x8(_mm512_castsi256_si512(q5_0), q5_1, 1);
+}
+
+// used for block_q6_K
+inline void bytes_from_nibbles_128(__m512i& r0, __m512i& r1, const uint8_t * qs, const uint8_t * qh) {
+    const __m256i m4 = _mm256_set1_epi8(0xF);
+    const __m256i m2 = _mm256_set1_epi8(0x3);
+
+    const __m256i q6bits1 = _mm256_loadu_si256((const __m256i *)qs);
+    const __m256i q6bits2 = _mm256_loadu_si256((const __m256i *)(qs + 32));
+    const __m256i q6bitsH = _mm256_loadu_si256((const __m256i *)qh);
+
+    const __m256i q6h_0 = _mm256_slli_epi16(_mm256_and_si256(                  q6bitsH,     m2), 4);
+    const __m256i q6h_1 = _mm256_slli_epi16(_mm256_and_si256(_mm256_srli_epi16(q6bitsH, 2), m2), 4);
+    const __m256i q6h_2 = _mm256_slli_epi16(_mm256_and_si256(_mm256_srli_epi16(q6bitsH, 4), m2), 4);
+    const __m256i q6h_3 = _mm256_slli_epi16(_mm256_and_si256(_mm256_srli_epi16(q6bitsH, 6), m2), 4);
+
+    const __m256i q6_0 = _mm256_or_si256(_mm256_and_si256(q6bits1, m4), q6h_0);
+    const __m256i q6_1 = _mm256_or_si256(_mm256_and_si256(q6bits2, m4), q6h_1);
+    const __m256i q6_2 = _mm256_or_si256(_mm256_and_si256(_mm256_srli_epi16(q6bits1, 4), m4), q6h_2);
+    const __m256i q6_3 = _mm256_or_si256(_mm256_and_si256(_mm256_srli_epi16(q6bits2, 4), m4), q6h_3);
+
+    r0 = _mm512_inserti32x8(_mm512_castsi256_si512(q6_0), q6_1, 1);
+    r1 = _mm512_inserti32x8(_mm512_castsi256_si512(q6_2), q6_3, 1);
+}
+
+inline __m512i packNibbles(__m512i r0, __m512i r1) {
+    return _mm512_or_si512(r0, _mm512_slli_epi16(r1, 4));
+}
+
+template 
+inline void pack_qs(void * RESTRICT packed_B, const TB * RESTRICT B, int KB) {
+    int8_t tmp[8 * 64];
+    __m256i v[8], v2[8];
+    for (int n = 0; n < 8; ++n) {
+        v[n] = bytes_from_nibbles_32(B[n * KB].qs);
+    }
+    transpose_8x8_32bit(v, v2);
+    for (int n = 0; n < 8; ++n) {
+        _mm256_storeu_si256((__m256i *)(tmp + n * 64), v2[n]);
+    }
+    for (int n = 0; n < 8; ++n) {
+        v[n] = bytes_from_nibbles_32(B[(n + 8) * KB].qs);
+    }
+    transpose_8x8_32bit(v, v2);
+    for (int n = 0; n < 8; ++n) {
+        _mm256_storeu_si256((__m256i *)(tmp + n * 64 + 32), v2[n]);
+    }
+
+    // pack again with 128 to fully utilize vector length
+    for (int n = 0; n < 8; n += 2) {
+        __m512i r0 = _mm512_loadu_si512((const __m512i *)(tmp + n * 64));
+        __m512i r1 = _mm512_loadu_si512((const __m512i *)(tmp + n * 64 + 64));
+        __m512i r1r0 = packNibbles(r0, r1);
+        _mm512_storeu_si512((__m512i *)((char *)packed_B + n * 32), r1r0);
+    }
+}
+
+template <>
+inline void pack_qs(void * RESTRICT packed_B, const block_q8_0 * RESTRICT B, int KB) {
+    __m256i v[8], v2[8];
+    for (int n = 0; n < 8; ++n) {
+        v[n] = _mm256_loadu_si256((const __m256i *)(B[n * KB].qs));
+    }
+    transpose_8x8_32bit(v, v2);
+    for (int n = 0; n < 8; ++n) {
+        _mm256_storeu_si256((__m256i *)((char *)packed_B + n * 64), v2[n]);
+    }
+    for (int n = 0; n < 8; ++n) {
+        v[n] = _mm256_loadu_si256((const __m256i *)(B[(n + 8) * KB].qs));
+    }
+    transpose_8x8_32bit(v, v2);
+    for (int n = 0; n < 8; ++n) {
+        _mm256_storeu_si256((__m256i *)((char *)packed_B + n * 64 + 32), v2[n]);
+    }
+}
+
+template <>
+inline void pack_qs(void * RESTRICT packed_B, const block_q4_K * RESTRICT B, int KB) {
+    __m512i v[16];
+    // QK_K 256 with 8 groups, handle 2 groups at a time
+    char * pb = (char *)packed_B;
+    for (int k = 0; k < QK_K / 64; ++k) {
+        // pack 2 groups { n, g,  k} to {g, k/4, 4n}
+        //          e.g. {16, 2, 32} to {2,   8, 64}
+        for (int n = 0; n < TILE_N; ++n) {
+            v[n] = bytes_from_nibbles_64(B[n * KB].qs + k * 32);
+        }
+
+        transpose_16x16_32bit(v);
+
+        // pack again with 128 to fully utilize vector length
+        for (int n = 0; n < TILE_N; n += 2) {
+            _mm512_storeu_si512((__m512i *)pb, packNibbles(v[n], v[n + 1]));
+            pb += 64;
+        }
+    }
+}
+
+template <>
+inline void pack_qs(void * RESTRICT packed_B, const block_q5_K * RESTRICT B, int KB) {
+    __m512i v[16];
+    const __m512i lowMask = _mm512_set1_epi8(0xF);
+    // QK_K 256 with 8 groups, handle 2 groups at a time
+    char * pb = (char *)packed_B;
+    char * ph = (char *)packed_B + (QK_K / 2) * TILE_N;
+    for (int k = 0; k < QK_K / 64; ++k) {
+        // pack 2 groups { n, g,  k} to {g, k/4, 4n}
+        //          e.g. {16, 2, 32} to {2,   8, 64}
+        for (int n = 0; n < TILE_N; ++n) {
+            v[n] = bytes_from_nibbles_64(B[n * KB].qs + k * 32, B[n * KB].qh, /* group */2 * k);
+        }
+
+        transpose_16x16_32bit(v);
+
+        // 1. pack lower 4bits with 2 groups
+        for (int n = 0; n < TILE_N; n += 2) {
+            // get lower 4 bits
+            const __m512i r0 = _mm512_and_si512(v[n], lowMask);
+            const __m512i r1 = _mm512_and_si512(v[n + 1], lowMask);
+            _mm512_storeu_si512((__m512i *)pb, packNibbles(r0, r1)); pb += 64;
+        }
+
+        // 2. pack higher 1bit with 2 groups
+        const __m512i hmask = _mm512_set1_epi8(0x10);
+        for (int g = 0; g < 2; ++g) {
+            __m512i hbits = _mm512_setzero_si512();
+            hbits = _mm512_add_epi8(hbits, _mm512_srli_epi16(_mm512_and_si512(v[g * 8 + 0], hmask), 4));
+            hbits = _mm512_add_epi8(hbits, _mm512_srli_epi16(_mm512_and_si512(v[g * 8 + 1], hmask), 3));
+            hbits = _mm512_add_epi8(hbits, _mm512_srli_epi16(_mm512_and_si512(v[g * 8 + 2], hmask), 2));
+            hbits = _mm512_add_epi8(hbits, _mm512_srli_epi16(_mm512_and_si512(v[g * 8 + 3], hmask), 1));
+            hbits = _mm512_add_epi8(hbits,                   _mm512_and_si512(v[g * 8 + 4], hmask)    );
+            hbits = _mm512_add_epi8(hbits, _mm512_slli_epi16(_mm512_and_si512(v[g * 8 + 5], hmask), 1));
+            hbits = _mm512_add_epi8(hbits, _mm512_slli_epi16(_mm512_and_si512(v[g * 8 + 6], hmask), 2));
+            hbits = _mm512_add_epi8(hbits, _mm512_slli_epi16(_mm512_and_si512(v[g * 8 + 7], hmask), 3));
+            _mm512_storeu_si512((__m512i *)ph, hbits); ph += 64;
+        }
+    }
+}
+
+template <>
+inline void pack_qs(void * RESTRICT packed_B, const block_q6_K * RESTRICT B, int KB) {
+    __m512i v[32];
+    const __m512i lowMask = _mm512_set1_epi8(0xF);
+    // QK_K 256 with 8 groups, handle 4 groups at a time
+    char * pb = (char *)packed_B;
+    char * ph = (char *)packed_B + (QK_K / 2) * TILE_N;
+    for (int k = 0; k < QK_K / 128; ++k) {
+        for (int n = 0; n < TILE_N; ++n) {
+            bytes_from_nibbles_128(v[n], v[n + 16], B[n * KB].ql + k * 64, B[n * KB].qh + k * 32);
+        }
+
+        // top half: group 0,1 or 4,5; bottom half: group 2,3 or 6,7
+        transpose_16x16_32bit(v);
+        transpose_16x16_32bit(v + 16);
+
+        // 1. pack lower 4bits with 4 groups
+        for (int n = 0; n < 32; n += 2) {
+            const __m512i r0 = _mm512_and_si512(v[n], lowMask);
+            const __m512i r1 = _mm512_and_si512(v[n + 1], lowMask);
+            _mm512_storeu_si512((__m512i *)pb, packNibbles(r0, r1)); pb += 64;
+        }
+
+        // 2. pack higher 2bit with 4 groups
+        const __m512i hmask = _mm512_set1_epi8(0x30);
+        for (int g = 0; g < 8; ++g) {
+            __m512i hbits = _mm512_setzero_si512();
+            hbits = _mm512_add_epi8(hbits, _mm512_srli_epi16(_mm512_and_si512(v[g * 4 + 0], hmask), 4));
+            hbits = _mm512_add_epi8(hbits, _mm512_srli_epi16(_mm512_and_si512(v[g * 4 + 1], hmask), 2));
+            hbits = _mm512_add_epi8(hbits,                   _mm512_and_si512(v[g * 4 + 2], hmask)    );
+            hbits = _mm512_add_epi8(hbits, _mm512_slli_epi16(_mm512_and_si512(v[g * 4 + 3], hmask), 2));
+            _mm512_storeu_si512((__m512i *)ph, hbits); ph += 64;
+        }
+    }
+}
+
+template <>
+inline void pack_qs(void * RESTRICT packed_B, const block_iq4_xs * RESTRICT B, int KB) {
+    __m512i v[16];
+    char * pb = (char *)packed_B;
+    for (int k = 0; k < QK_K / 64; ++k) {
+        for (int n = 0; n < TILE_N; ++n) {
+            __m256i r0 = bytes_from_nibbles_32(B[n * KB].qs + k * 32 +  0);
+            __m256i r1 = bytes_from_nibbles_32(B[n * KB].qs + k * 32 + 16);
+            v[n] = _mm512_inserti32x8(_mm512_castsi256_si512(r0), r1, 1);
+        }
+
+        transpose_16x16_32bit(v);
+
+        // pack again with 128 to fully utilize vector length
+        for (int n = 0; n < TILE_N; n += 2) {
+            _mm512_storeu_si512((__m512i *)pb, packNibbles(v[n], v[n + 1]));
+            pb += 64;
+        }
+    }
+}
+
+// pack B to vnni formats in 4bits or 8 bits
+void pack_B(void * RESTRICT packed_B, const block_q4_0 * RESTRICT B, int KB) {
+    pack_qs(packed_B, B, KB);
+    ggml_half * d0 = reinterpret_cast((char *)packed_B + TILE_N * TILE_K / 2);
+    for (int n = 0; n < TILE_N; ++n) {
+        d0[n] = B[n * KB].d;
+    }
+}
+
+void pack_B(void * RESTRICT packed_B, const block_q4_1 * RESTRICT B, int KB) {
+    pack_qs(packed_B, B, KB);
+    ggml_half * d0 = reinterpret_cast((char *)packed_B + TILE_N * TILE_K / 2);
+    ggml_half * m0 = d0 + TILE_N;
+    for (int n = 0; n < TILE_N; ++n) {
+        d0[n] = B[n * KB].d;
+        m0[n] = B[n * KB].m;
+    }
+}
+
+inline void s8s8_compensation(void * RESTRICT packed_B) {
+    // packed_B layout:
+    //   quants {TILE_N, TILEK}  int8_t
+    //   d0     {TILE_N}      ggml_half
+    //   comp   {TILE_N}        int32_t
+    const int offset = TILE_N * TILE_K + TILE_N * sizeof(ggml_half);
+    __m512i vcomp = _mm512_setzero_si512();
+    const __m512i off = _mm512_set1_epi8(static_cast(0x80));
+    for (int k = 0; k < 8; ++k) {
+        __m512i vb = _mm512_loadu_si512((const __m512i *)((const char *)packed_B + k * 64));
+        vcomp = _mm512_dpbusd_epi32(vcomp, off, vb);
+    }
+    _mm512_storeu_si512((__m512i *)((char *)(packed_B) + offset), vcomp);
+}
+
+void pack_B(void * RESTRICT packed_B, const block_q8_0 * RESTRICT B, int KB) {
+    pack_qs(packed_B, B, KB);
+    ggml_half * d0 = reinterpret_cast((char *)packed_B + TILE_N * TILE_K);
+    for (int n = 0; n < TILE_N; ++n) {
+        d0[n] = B[n * KB].d;
+    }
+    s8s8_compensation(packed_B);
+}
+
+// convert 8 * {min, scale} from int6 to int8
+inline void unpack_mins_and_scales(const uint8_t * scales, uint32_t * utmp) {
+    const uint32_t kmask1 = 0x3f3f3f3f;
+    const uint32_t kmask2 = 0x0f0f0f0f;
+    const uint32_t kmask3 = 0x03030303;
+
+    memcpy(utmp, scales, 12);
+    utmp[3] = ((utmp[2] >> 4) & kmask2) | (((utmp[1] >> 6) & kmask3) << 4);
+    const uint32_t uaux = utmp[1] & kmask1;
+    utmp[1] = (utmp[2] & kmask2) | (((utmp[0] >> 6) & kmask3) << 4);
+    utmp[2] = uaux;
+    utmp[0] &= kmask1;
+}
+
+// packed_B layout:
+//   quants {8, TILE_N, 16}  uint8
+//   scales {8, TILE_N}      uint8
+//   mins   {8, TILE_N}      uint8
+//   d      {TILE_N}     ggml_half
+//   dmin   {TILE_N}     ggml_half
+void pack_B(void * RESTRICT packed_B, const block_q4_K * RESTRICT B, int KB) {
+    pack_qs(packed_B, B, KB);
+
+    uint8_t * scales = reinterpret_cast((char *)packed_B + (QK_K / 2) * TILE_N);
+    uint8_t * mins = scales + 8 * TILE_N;
+    ggml_half * d = reinterpret_cast(mins + 8 * TILE_N);
+    ggml_half * dmin = d + TILE_N;
+
+    union {
+        uint32_t u32[4];
+        uint8_t  u8[16];
+    } s;
+
+    for (int n = 0; n < TILE_N; ++n) {
+        unpack_mins_and_scales(B[n * KB].scales, s.u32);
+        for (int k = 0; k < 8; ++k) {
+            scales[k * TILE_N + n] = s.u8[k];
+            mins[(k >> 1) * TILE_N * 2 + n * 2 + (k & 0x1)] = s.u8[k + 8];
+        }
+        d[n] = B[n * KB].d;
+        dmin[n] = B[n * KB].dmin;
+    }
+}
+
+// packed_B layout:
+//   quants {8, TILE_N, 16}  uint8
+//   qh     {8, TILE_N,  4}  uint8
+//   scales {8, TILE_N}      uint8
+//   mins   {8, TILE_N}      uint8
+//   d      {TILE_N}     ggml_half
+//   dmin   {TILE_N}     ggml_half
+void pack_B(void * RESTRICT packed_B, const block_q5_K * RESTRICT B, int KB) {
+    pack_qs(packed_B, B, KB);
+
+    uint8_t * scales = reinterpret_cast((char *)packed_B + (QK_K / 2) * TILE_N + (QK_K / 8) * TILE_N);
+    uint8_t * mins = scales + 8 * TILE_N;
+    ggml_half * d = reinterpret_cast(mins + 8 * TILE_N);
+    ggml_half * dmin = d + TILE_N;
+
+    union {
+        uint32_t u32[4];
+        uint8_t  u8[16];
+    } s;
+
+    for (int n = 0; n < TILE_N; ++n) {
+        unpack_mins_and_scales(B[n * KB].scales, s.u32);
+        for (int k = 0; k < 8; ++k) {
+            scales[k * TILE_N + n] = s.u8[k];
+            mins[(k >> 1) * TILE_N * 2 + n * 2 + (k & 0x1)] = s.u8[k + 8];
+        }
+        d[n] = B[n * KB].d;
+        dmin[n] = B[n * KB].dmin;
+    }
+}
+
+// packed_B layout:
+//   quants {16, TILE_N, 8}  uint8
+//   qh     {16, TILE_N, 4}  uint8
+//   scales {16, TILE_N}      uint8
+//   d      {TILE_N}     ggml_half
+void pack_B(void * RESTRICT packed_B, const block_q6_K * RESTRICT B, int KB) {
+    pack_qs(packed_B, B, KB);
+
+    uint8_t * scales = reinterpret_cast((char *)packed_B + (QK_K / 2) * TILE_N + (QK_K / 4) * TILE_N);
+    ggml_half * d = reinterpret_cast(scales + 16 * TILE_N);
+    for (int n = 0; n < TILE_N; ++n) {
+        const int8_t * ps = B[n * KB].scales;
+        for (int k = 0; k < 16; ++k) {
+            scales[k * TILE_N + n] = ps[k];
+        }
+        d[n] = B[n * KB].d;
+    }
+}
+
+// packed_B layout:
+//   quants {8, TILE_N, 16}  uint8
+//   scales {8, TILE_N}       int8
+//   d      {TILE_N}     ggml_half
+void pack_B(void * RESTRICT packed_B, const block_iq4_xs * RESTRICT B, int KB) {
+    pack_qs(packed_B, B, KB);
+
+    int8_t * scales = reinterpret_cast((char *)packed_B + (QK_K / 2) * TILE_N);
+    ggml_half * d = reinterpret_cast(scales + 8 * TILE_N);
+
+    // pack the scales
+    for (int n = 0; n < TILE_N; ++n) {
+        uint16_t sh = B[n * KB].scales_h;
+        for (int k = 0; k < 8; k += 2) {
+            const int16_t ls1 = ((B[n * KB].scales_l[k / 2] & 0xf) | ((sh << 4) & 0x30)) - 32;
+            const int16_t ls2 = ((B[n * KB].scales_l[k / 2] >>  4) | ((sh << 2) & 0x30)) - 32;
+            scales[(k + 0) * TILE_N + n] = ls1;
+            scales[(k + 1) * TILE_N + n] = ls2;
+            sh >>= 4;
+        }
+        d[n] = B[n * KB].d;
+    }
+}
+
+template>
+void unpack_B(packed_B_t * RESTRICT tile, const void * RESTRICT packed_B) {
+    GGML_UNUSED(tile);
+    GGML_UNUSED(packed_B);
+};
+
+template <>
+void unpack_B(int8_t * RESTRICT tile, const void * RESTRICT packed_B) {
+  const __m512i off = _mm512_set1_epi8(8);
+  const __m512i lowMask = _mm512_set1_epi8(0xF);
+  for (int n = 0; n < 8; n += 2) {
+    __m512i bytes = _mm512_loadu_si512((const __m512i *)((const char *)packed_B + n * 32));
+    const __m512i r0 = _mm512_sub_epi8(_mm512_and_si512(bytes, lowMask), off);
+    const __m512i r1 = _mm512_sub_epi8(_mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask), off);
+    _mm512_storeu_si512((__m512i *)(tile + n * 64 +  0), r0);
+    _mm512_storeu_si512((__m512i *)(tile + n * 64 + 64), r1);
+  }
+}
+
+template <>
+void unpack_B(uint8_t * RESTRICT tile, const void * RESTRICT packed_B) {
+    const __m512i lowMask = _mm512_set1_epi8(0xF);
+    for (int n = 0; n < 8; n += 2) {
+        __m512i bytes = _mm512_loadu_si512((const __m512i *)((const char *)packed_B + n * 32));
+        const __m512i r0 = _mm512_and_si512(bytes, lowMask);
+        const __m512i r1 = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+        _mm512_storeu_si512((__m512i *)(tile + n * 64 +  0), r0);
+        _mm512_storeu_si512((__m512i *)(tile + n * 64 + 64), r1);
+    }
+}
+
+// packed_B_t for QKK is int8_t
+template 
+void unpack_B(int8_t * RESTRICT tile, const void * RESTRICT packed_B, int k) {
+    const int packed_B_group_size = QK_K / 2 * TILE_N / 8;
+    const char * packed_B_group = (const char *)packed_B + k * packed_B_group_size;
+    const __m512i lowMask = _mm512_set1_epi8(0xF);
+    for (int n = 0; n < 8; n += 2) {
+        __m512i bytes = _mm512_loadu_si512(packed_B_group + n * 32);
+        const __m512i r0 = _mm512_and_si512(bytes, lowMask);
+        const __m512i r1 = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+        _mm512_storeu_si512((__m512i *)(tile + n * 64 +  0), r0);
+        _mm512_storeu_si512((__m512i *)(tile + n * 64 + 64), r1);
+    }
+}
+
+template <>
+void unpack_B(int8_t * RESTRICT tile, const void * RESTRICT packed_B, int k) {
+    // lower 4bits, stride 256 bytes
+    const int packed_l4_group_size = QK_K / 2 * TILE_N / 8;
+    const char * pb = (const char *)packed_B + k * packed_l4_group_size;
+
+    // higher 1bit, stride 64 bytes
+    const int packed_h1_group_size = QK_K / 8 * TILE_N / 8;
+    const char * ph = (const char *)packed_B + (QK_K / 2) * TILE_N + k * packed_h1_group_size;
+    const __m512i hbits = _mm512_loadu_si512(ph);
+
+    const __m512i lowMask = _mm512_set1_epi8(0xF);
+    __m512i hmask0 = _mm512_set1_epi8(0x1);
+    __m512i hmask1 = _mm512_set1_epi8(0x2);
+
+    for (int n = 0; n < 8; n += 2) {
+        __m512i bytes = _mm512_loadu_si512(pb + n * 32);
+        __m512i r0 = _mm512_and_si512(bytes, lowMask);
+        __m512i r1 = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+        __m512i h0 = _mm512_slli_epi16(_mm512_srli_epi16(_mm512_and_si512(hbits, hmask0), n), 4);
+        __m512i h1 = _mm512_slli_epi16(_mm512_srli_epi16(_mm512_and_si512(hbits, hmask1), n + 1), 4);
+
+        hmask0 = _mm512_slli_epi16(hmask0, 2);
+        hmask1 = _mm512_slli_epi16(hmask1, 2);
+        r0 = _mm512_add_epi8(r0, h0);
+        r1 = _mm512_add_epi8(r1, h1);
+        _mm512_storeu_si512((__m512i *)(tile + n * 64 +  0), r0);
+        _mm512_storeu_si512((__m512i *)(tile + n * 64 + 64), r1);
+    }
+}
+
+template <>
+void unpack_B(int8_t * RESTRICT tile, const void * RESTRICT packed_B, int k) {
+    // lower 4bits, stride 128 bytes
+    const int packed_l4_group_size = QK_K / 2 * TILE_N / 16;
+    const char * pb = (const char *)packed_B + k * packed_l4_group_size;
+
+    // higher 2bits, stride 64 bytes
+    const int packed_h2_group_size = QK_K / 4 * TILE_N / 16;
+    const char * ph = (const char *)packed_B + (QK_K / 2) * TILE_N + k * packed_h2_group_size;
+    const __m512i hbits = _mm512_loadu_si512(ph);
+
+    const __m512i off = _mm512_set1_epi8(32);
+    const __m512i lowMask = _mm512_set1_epi8(0xF);
+    __m512i hmask0 = _mm512_set1_epi8(0x3); // 0011
+    __m512i hmask1 = _mm512_set1_epi8(0xC); // 1100
+
+    // notes: skip zero padding from row4 to row7 as we have done so in `unpack_A`
+    __m512i bytes = _mm512_loadu_si512(pb);
+    __m512i r0 = _mm512_and_si512(bytes, lowMask);
+    __m512i r1 = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+    __m512i h0 = _mm512_slli_epi16(_mm512_and_si512(hbits, hmask0), 4);
+    __m512i h1 = _mm512_slli_epi16(_mm512_and_si512(hbits, hmask1), 2);
+    _mm512_storeu_si512((__m512i *)(tile +  0), _mm512_sub_epi8(_mm512_add_epi8(r0, h0), off));
+    _mm512_storeu_si512((__m512i *)(tile + 64), _mm512_sub_epi8(_mm512_add_epi8(r1, h1), off));
+
+    hmask0 = _mm512_slli_epi16(hmask0, 4);
+    hmask1 = _mm512_slli_epi16(hmask1, 4);
+
+    bytes = _mm512_loadu_si512(pb + 64);
+    r0 = _mm512_and_si512(bytes, lowMask);
+    r1 = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+    h0 =                   _mm512_and_si512(hbits, hmask0);
+    h1 = _mm512_srli_epi16(_mm512_and_si512(hbits, hmask1), 2);
+    _mm512_storeu_si512((__m512i *)(tile + 128), _mm512_sub_epi8(_mm512_add_epi8(r0, h0), off));
+    _mm512_storeu_si512((__m512i *)(tile + 192), _mm512_sub_epi8(_mm512_add_epi8(r1, h1), off));
+}
+
+template <>
+void unpack_B(int8_t * RESTRICT tile, const void * RESTRICT packed_B, int k) {
+    static const __m512i values128 = _mm512_set_epi8(
+        113, 89, 69, 53, 38, 25, 13, 1, -10, -22, -35, -49, -65, -83, -104, -127,
+        113, 89, 69, 53, 38, 25, 13, 1, -10, -22, -35, -49, -65, -83, -104, -127,
+        113, 89, 69, 53, 38, 25, 13, 1, -10, -22, -35, -49, -65, -83, -104, -127,
+        113, 89, 69, 53, 38, 25, 13, 1, -10, -22, -35, -49, -65, -83, -104, -127
+    );
+
+    const int packed_B_group_size = QK_K / 2 * TILE_N / 8;
+    const char * pb = (const char *)packed_B + k * packed_B_group_size;
+    const __m512i lowMask = _mm512_set1_epi8(0xF);
+
+    for (int n = 0; n < 8; n += 2) {
+        __m512i bytes = _mm512_loadu_si512(pb + n * 32);
+        const __m512i r0 = _mm512_shuffle_epi8(values128, _mm512_and_si512(bytes, lowMask));
+        const __m512i r1 = _mm512_shuffle_epi8(values128, _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask));
+        _mm512_storeu_si512((__m512i *)(tile + n * 64 +  0), r0);
+        _mm512_storeu_si512((__m512i *)(tile + n * 64 + 64), r1);
+    }
+}
+
+template 
+struct acc_C {};
+
+template 
+struct acc_C {
+    static void apply(float * RESTRICT C, int ldc, const int32_t * RESTRICT tile, const block_q8_0 * A, int lda, const void * packed_B, int nr) {
+        const int offset = TILE_N * TILE_K / 2;
+        const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)((const char *)packed_B + offset)));
+
+        for (int m = 0; m < nr; ++m) {
+            const __m512 vd1 = _mm512_set1_ps(GGML_FP16_TO_FP32(A[m * lda].d));
+            const __m512 vtile = _mm512_cvtepi32_ps(_mm512_loadu_si512(tile + m * TILE_N));
+
+            __m512 vsum;
+            if (is_acc) {
+                vsum = _mm512_loadu_ps(C + m * ldc);
+            } else {
+                vsum = _mm512_set1_ps(0.f);
+            }
+            vsum = _mm512_fmadd_ps(vtile, _mm512_mul_ps(vd0, vd1), vsum);
+            _mm512_storeu_ps(C + m * ldc, vsum);
+        }
+    }
+};
+
+template 
+struct acc_C {
+    static void apply(float * RESTRICT C, int ldc, const int32_t * RESTRICT tile, const block_q8_1 * A, int lda, const void * packed_B, int nr) {
+        const int offset = TILE_N * TILE_K / 2;
+        const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)((const char *)packed_B + offset)));
+        const __m512 vm0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)((const char *)packed_B + offset + TILE_N * sizeof(ggml_half))));
+
+        for (int m = 0; m < nr; ++m) {
+            const __m512 vd1 = _mm512_set1_ps(GGML_FP16_TO_FP32(A[m * lda].d));
+            const __m512 vs1 = _mm512_set1_ps(GGML_FP16_TO_FP32(A[m * lda].s));
+            const __m512 vtile = _mm512_cvtepi32_ps(_mm512_loadu_si512(tile + m * TILE_N));
+
+            __m512 vsum;
+            if (is_acc) {
+                vsum = _mm512_loadu_ps(C + m * ldc);
+            } else {
+                vsum = _mm512_set1_ps(0.f);
+            }
+            vsum = _mm512_fmadd_ps(vtile, _mm512_mul_ps(vd0, vd1), vsum);
+            vsum = _mm512_fmadd_ps(vm0, vs1, vsum);
+            _mm512_storeu_ps(C + m * ldc, vsum);
+        }
+    }
+};
+
+template 
+struct acc_C {
+    static void apply(float * RESTRICT C, int ldc, const int32_t * RESTRICT tile, const block_q8_0 * A, int lda, const void * packed_B, int nr) {
+        const int offset = TILE_N * TILE_K;
+        const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)((const char *)packed_B + offset)));
+
+        for (int m = 0; m < nr; ++m) {
+            const __m512 vd1 = _mm512_set1_ps(GGML_FP16_TO_FP32(A[m * lda].d));
+            const __m512 vtile = _mm512_cvtepi32_ps(_mm512_loadu_si512(tile + m * TILE_N));
+
+            __m512 vsum;
+            if (is_acc) {
+                vsum = _mm512_loadu_ps(C + m * ldc);
+            } else {
+                vsum = _mm512_set1_ps(0.f);
+            }
+            vsum = _mm512_fmadd_ps(vtile, _mm512_mul_ps(vd0, vd1), vsum);
+            _mm512_storeu_ps(C + m * ldc, vsum);
+        }
+    }
+};
+
+template 
+struct acc_C {
+    static void apply(float * RESTRICT C, int ldc, const int32_t * RESTRICT tile, const block_q8_K * A, int lda, const void * packed_B, int nr) {
+        const uint8_t * scales = reinterpret_cast((const char *)packed_B + (QK_K / 2) * TILE_N);
+        const uint8_t * mins = scales + 8 * TILE_N;
+        const ggml_half * d0 = reinterpret_cast(mins + 8 * TILE_N);
+        const ggml_half * dmin = d0 + TILE_N;
+
+        const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)d0));
+        const __m512 vdmin = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)dmin));
+
+        for (int m = 0; m < nr; ++m) {
+            const float d1 = A[m * lda].d;
+            const __m512 vd = _mm512_mul_ps(_mm512_set1_ps(d1), vd0);
+            const __m512 vdm = _mm512_mul_ps(_mm512_set1_ps(-d1), vdmin);
+            const __m512 vtile = _mm512_cvtepi32_ps(_mm512_loadu_si512(tile + m * TILE_N));
+
+            __m512 vsum;
+            if (is_acc) {
+                vsum = _mm512_loadu_ps(C + m * ldc);
+            } else {
+                vsum = _mm512_set1_ps(0.f);
+            }
+
+            const __m256i q8sums = _mm256_loadu_si256((const __m256i *)A[m * lda].bsums);
+            const __m128i q8s = _mm_hadd_epi16(_mm256_extracti128_si256(q8sums, 0), _mm256_extracti128_si256(q8sums, 1));
+
+            __m512i acc_m = _mm512_setzero_si512();
+            for (int k = 0; k < 4; ++k) {
+                __m512i vmask = _mm512_set1_epi32(k);
+                __m512i va = _mm512_permutexvar_epi32(vmask, _mm512_castsi128_si512(q8s));
+                __m512i vb = _mm512_cvtepi8_epi16(_mm256_loadu_si256((const __m256i *)(mins + k * 32)));
+                acc_m = _mm512_dpwssds_epi32(acc_m, va, vb);
+            }
+
+            vsum = _mm512_fmadd_ps(vtile, vd, vsum);
+            vsum = _mm512_fmadd_ps(_mm512_cvtepi32_ps(acc_m), vdm, vsum);
+            _mm512_storeu_ps(C + m * ldc, vsum);
+        }
+    }
+};
+
+template 
+struct acc_C {
+    static void apply(float * RESTRICT C, int ldc, const int32_t * RESTRICT tile, const block_q8_K * A, int lda, const void * packed_B, int nr) {
+        const uint8_t * scales = reinterpret_cast((const char *)packed_B + (QK_K / 2) * TILE_N + (QK_K / 8) * TILE_N);
+        const uint8_t * mins = scales + 8 * TILE_N;
+        const ggml_half * d0 = reinterpret_cast(mins + 8 * TILE_N);
+        const ggml_half * dmin = d0 + TILE_N;
+
+        const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)d0));
+        const __m512 vdmin = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)dmin));
+
+        for (int m = 0; m < nr; ++m) {
+            const float d1 = A[m * lda].d;
+            const __m512 vd = _mm512_mul_ps(_mm512_set1_ps(d1), vd0);
+            const __m512 vdm = _mm512_mul_ps(_mm512_set1_ps(-d1), vdmin);
+            const __m512 vtile = _mm512_cvtepi32_ps(_mm512_loadu_si512(tile + m * TILE_N));
+
+            __m512 vsum;
+            if (is_acc) {
+                vsum = _mm512_loadu_ps(C + m * ldc);
+            } else {
+                vsum = _mm512_set1_ps(0.f);
+            }
+
+            const __m256i q8sums = _mm256_loadu_si256((const __m256i *)A[m * lda].bsums);
+            const __m128i q8s = _mm_hadd_epi16(_mm256_extracti128_si256(q8sums, 0), _mm256_extracti128_si256(q8sums, 1));
+
+            __m512i acc_m = _mm512_setzero_si512();
+            for (int k = 0; k < 4; ++k) {
+                __m512i vmask = _mm512_set1_epi32(k);
+                __m512i va = _mm512_permutexvar_epi32(vmask, _mm512_castsi128_si512(q8s));
+                __m512i vb = _mm512_cvtepi8_epi16(_mm256_loadu_si256((const __m256i *)(mins + k * 32)));
+                acc_m = _mm512_dpwssds_epi32(acc_m, va, vb);
+            }
+
+            vsum = _mm512_fmadd_ps(vtile, vd, vsum);
+            vsum = _mm512_fmadd_ps(_mm512_cvtepi32_ps(acc_m), vdm, vsum);
+            _mm512_storeu_ps(C + m * ldc, vsum);
+        }
+    }
+};
+
+template 
+struct acc_C {
+    static void apply(float * RESTRICT C, int ldc, const int32_t * RESTRICT tile, const block_q8_K * A, int lda, const void * packed_B, int nr) {
+        const uint8_t * scales = reinterpret_cast((const char *)packed_B + (QK_K / 2) * TILE_N + (QK_K / 4) * TILE_N);
+        const ggml_half * d0 = reinterpret_cast(scales + 16 * TILE_N);
+
+        const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)d0));
+
+        for (int m = 0; m < nr; ++m) {
+            const float d1 = A[m * lda].d;
+            const __m512 vd = _mm512_mul_ps(_mm512_set1_ps(d1), vd0);
+            const __m512 vtile = _mm512_cvtepi32_ps(_mm512_loadu_si512(tile + m * TILE_N));
+
+            __m512 vsum;
+            if (is_acc) {
+                vsum = _mm512_loadu_ps(C + m * ldc);
+            } else {
+                vsum = _mm512_set1_ps(0.f);
+            }
+
+            vsum = _mm512_fmadd_ps(vtile, vd, vsum);
+            _mm512_storeu_ps(C + m * ldc, vsum);
+        }
+    }
+};
+
+template 
+struct acc_C {
+    static void apply(float * RESTRICT C, int ldc, const int32_t * RESTRICT tile, const block_q8_K * A, int lda, const void * packed_B, int nr) {
+        const int8_t * scales = reinterpret_cast((const char *)packed_B + (QK_K / 2) * TILE_N);
+        const ggml_half * d0 = reinterpret_cast(scales + 8 * TILE_N);
+
+        const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)d0));
+
+        for (int m = 0; m < nr; ++m) {
+            const float d1 = A[m * lda].d;
+            const __m512 vd = _mm512_mul_ps(_mm512_set1_ps(d1), vd0);
+            const __m512 vtile = _mm512_cvtepi32_ps(_mm512_loadu_si512(tile + m * TILE_N));
+
+            __m512 vsum;
+            if (is_acc) {
+                vsum = _mm512_loadu_ps(C + m * ldc);
+            } else {
+                vsum = _mm512_set1_ps(0.f);
+            }
+
+            vsum = _mm512_fmadd_ps(vtile, vd, vsum);
+            _mm512_storeu_ps(C + m * ldc, vsum);
+        }
+    }
+};
+
+template  constexpr int get_quants_size();
+template <> constexpr int get_quants_size() { return (QK_K / 2) * TILE_N; }
+template <> constexpr int get_quants_size() { return (QK_K / 2) * TILE_N + (QK_K / 8) * TILE_N; }
+template <> constexpr int get_quants_size() { return (QK_K / 2) * TILE_N + (QK_K / 4) * TILE_N; }
+template <> constexpr int get_quants_size() { return (QK_K / 2) * TILE_N; }
+
+// used for QKK format
+template ::value, int>::type = 0>
+inline void scale_C(const int32_t * RESTRICT tile, int32_t * RESTRICT sumi, const void * packed_B, int k, int nr) {
+    const uint8_t * scales = reinterpret_cast((const char *)packed_B + get_quants_size());
+    const __m512i vscale = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i *)(scales + k * TILE_N)));
+
+    for (int m = 0; m < nr; ++m) {
+        __m512i vsumi;
+        if (is_acc) {
+            vsumi = _mm512_loadu_si512(sumi + m * TILE_N);
+        } else {
+            vsumi = _mm512_setzero_si512();
+        }
+        __m512i vtile = _mm512_loadu_si512(tile + m * TILE_N);
+        vsumi = _mm512_add_epi32(vsumi, _mm512_mullo_epi32(vtile, vscale));
+        _mm512_storeu_si512((__m512i *)(sumi + m * TILE_N), vsumi);
+    }
+}
+
+template 
+struct tinygemm_kernel_avx {
+    static void apply(int K, const TA * RESTRICT A, const TB * RESTRICT B, TC * RESTRICT C, int ldc) {
+        GGML_UNUSED(K);
+        GGML_UNUSED(A);
+        GGML_UNUSED(B);
+        GGML_UNUSED(C);
+        GGML_UNUSED(ldc);
+    }
+};
+
+template 
+struct tinygemm_kernel_avx {
+    static void apply(int K, const float * RESTRICT A, const ggml_fp16_t * RESTRICT B, float * RESTRICT C, int ldc) {
+        constexpr int ROWS = BLOCK_M;
+        constexpr int COLS = BLOCK_N;
+        assert(BLOCK_K == 16);
+
+        __m512 va;
+        __m512 vb[COLS];
+        __m512 vc[ROWS * COLS];
+
+        auto loadc = [&](int idx) {
+            vc[idx] = _mm512_setzero_ps();
+        };
+        Unroll{}(loadc);
+
+        auto compute = [&](int idx, int k) {
+            // TODO: use `constexpr` here to get rid of interger div
+            // when upgraded to C++17
+            const int row = idx / COLS;
+            const int col = idx % COLS;
+
+            if (col == 0) {
+                va = _mm512_loadu_ps(A + row * K + k);
+            }
+            if (row == 0) {
+                vb[col] =  _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(B + col * K + k)));
+            }
+            vc[idx] = _mm512_fmadd_ps(va, vb[col], vc[idx]);
+        };
+
+        for (int k = 0; k < K; k += 16) {
+            Unroll{}(compute, k);
+        }
+
+        auto storec = [&](int idx) {
+            const int row = idx / COLS;
+            const int col = idx % COLS;
+            C[row * ldc + col] = _mm512_reduce_add_ps(vc[idx]);
+        };
+        Unroll{}(storec);
+    }
+};
+
+#define LAUNCH_TINYGEMM_KERNEL_AVX(MB_SIZE, NB_SIZE)                                \
+    tinygemm_kernel_avx::apply(    \
+        K, (const float *)src1->data + mb_start * K,                                \
+        (const type *)src0->data + nb_start * K,                                    \
+        (float *)dst->data + mb_start * ldc + nb_start, ldc);
+
+
+// re-organize in the format {NB, KB, TILE_SIZE}:
+#define PACKED_INDEX(n, k, KB, tile_size) (n * KB + k) * tile_size
+
+template
+void convert_B_packed_format(void * RESTRICT packed_B, const TB * RESTRICT B, int N, int K, int n_threads) {
+    const int NB = N / TILE_N;
+    const int KB = K / BLOCK_K;
+    const int TILE_SIZE = get_tile_size();
+
+    // parallel on NB should be enough
+    parallel_for(n_threads, NB, [&](int begin, int end) {
+        for (int n = begin; n < end; ++n) {
+            for (int k = 0; k < KB; ++k) {
+                int n0 = n * TILE_N;
+                pack_B((char *)packed_B + PACKED_INDEX(n, k, KB, TILE_SIZE), &B[n0 * KB + k], KB);
+            }
+        }
+    });
+}
+
+template 
+struct tinygemm_kernel_vnni {};
+
+template 
+struct tinygemm_kernel_vnni {
+    static void apply(int KB, const void * RESTRICT _A, const void * RESTRICT _B, float * RESTRICT C, int ldc) {
+
+        constexpr int COLS = BLOCK_N / 16;
+        const int TILE_SIZE = TILE_N * sizeof(block_q4_0);
+
+        const block_q8_0 * RESTRICT A = static_cast(_A);
+        const char * RESTRICT B = static_cast(_B);
+
+        __m512i va[8];
+        __m512 vc[COLS];
+        __m512 vd1;
+
+        // sum of offsets, shared across COLS
+        //
+        // avx512-vnni does not have `_mm512_dpbssd_epi32`,
+        // need to transfrom ss to us:
+        //   a * (b - 8) is equavilent to b * a - 8 * a
+        //   s    u   u                   u   s   u   s
+        //
+        __m512i vcomp;
+
+        const __m512i off = _mm512_set1_epi8(8);
+        const __m512i lowMask = _mm512_set1_epi8(0xF);
+
+        auto loadc = [&](int col) {
+            vc[col] = _mm512_setzero_ps();
+        };
+        Unroll{}(loadc);
+
+        auto compute = [&](int col, int i) {
+            // load a and compute compensation
+            if (col == 0) {
+                const int32_t * a_ptr = reinterpret_cast(A[0 * KB + i].qs);
+                vcomp = _mm512_setzero_si512();
+                for (int k = 0; k < 8; ++k) {
+                    va[k] = _mm512_set1_epi32(a_ptr[k]);
+                    vcomp = _mm512_dpbusd_epi32(vcomp, off, va[k]);
+                }
+                vd1 = _mm512_set1_ps(GGML_FP16_TO_FP32(A[0 * KB + i].d));
+            }
+
+            // load b
+            __m512i vsum = _mm512_setzero_si512();
+            const char * b_ptr = B + PACKED_INDEX(col, i, KB, TILE_SIZE);
+            for (int k = 0; k < 8; k += 2) {
+                __m512i bytes = _mm512_loadu_si512((const __m512i *)(b_ptr + k * 32));
+                __m512i vb0 = _mm512_and_si512(bytes, lowMask);
+                vsum = _mm512_dpbusd_epi32(vsum, vb0, va[k + 0]);
+                __m512i vb1 = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+                vsum = _mm512_dpbusd_epi32(vsum, vb1, va[k + 1]);
+            }
+            const int offset = TILE_N * TILE_K / 2;
+            const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(b_ptr + offset)));
+            vsum = _mm512_sub_epi32(vsum, vcomp);
+
+            vc[col] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(vsum), _mm512_mul_ps(vd0, vd1), vc[col]);
+        };
+
+        for (int i = 0; i < KB; ++i) {
+            Unroll{}(compute, i);
+        }
+
+        //store to C
+        auto storec = [&](int col) {
+            _mm512_storeu_ps((__m512i*)(C + 0 * ldc + col * 16), vc[col]);
+        };
+        Unroll{}(storec);
+    }
+};
+
+template 
+struct tinygemm_kernel_vnni {
+    static void apply(int KB, const void * RESTRICT _A, const void * RESTRICT _B, float * RESTRICT C, int ldc) {
+
+        constexpr int COLS = BLOCK_N / 16;
+        const int TILE_SIZE = TILE_N * sizeof(block_q4_1);
+
+        const block_q8_1 * RESTRICT A = static_cast(_A);
+        const char * RESTRICT B = static_cast(_B);
+
+        __m512i va[8];
+        __m512i vb[8];
+        __m512 vc[COLS];
+        __m512 vd1, vs1;
+
+        const __m512i lowMask = _mm512_set1_epi8(0xF);
+
+        auto loadc = [&](int col) {
+            vc[col] = _mm512_setzero_ps();
+        };
+        Unroll{}(loadc);
+
+        auto compute = [&](int col, int i) {
+            // load a
+            if (col == 0) {
+                const int32_t * a_ptr = reinterpret_cast(A[0 * KB + i].qs);
+                for (int k = 0; k < 8; ++k) {
+                    va[k] = _mm512_set1_epi32(a_ptr[k]);
+                }
+                vd1 = _mm512_set1_ps(GGML_FP16_TO_FP32(A[0 * KB + i].d));
+                vs1 = _mm512_set1_ps(GGML_FP16_TO_FP32(A[0 * KB + i].s));
+            }
+
+            // load b
+            const char * b_ptr = B + PACKED_INDEX(col, i, KB, TILE_SIZE);
+            for (int k = 0; k < 8; k += 2) {
+                __m512i bytes = _mm512_loadu_si512((const __m512i *)(b_ptr + k * 32));
+                vb[k + 0] = _mm512_and_si512(bytes, lowMask);
+                vb[k + 1] = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+            }
+            const int offset = TILE_N * TILE_K / 2;
+            const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(b_ptr + offset)));
+            const __m512 vm0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(b_ptr + offset + TILE_N * sizeof(ggml_half))));
+
+            __m512i vsum = _mm512_setzero_si512();
+            for (int k = 0; k < 8; ++k) {
+                vsum = _mm512_dpbusd_epi32(vsum, vb[k], va[k]);
+            }
+
+            vc[col] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(vsum), _mm512_mul_ps(vd0, vd1), vc[col]);
+            vc[col] = _mm512_fmadd_ps(vm0, vs1, vc[col]);
+        };
+
+        for (int i = 0; i < KB; ++i) {
+            Unroll{}(compute, i);
+        }
+
+        //store to C
+        auto storec = [&](int col) {
+            _mm512_storeu_ps((__m512i*)(C + 0 * ldc + col * 16), vc[col]);
+        };
+        Unroll{}(storec);
+    }
+};
+
+template 
+struct tinygemm_kernel_vnni {
+    static void apply(int KB, const void * RESTRICT _A, const void * RESTRICT _B, float * RESTRICT C, int ldc) {
+
+        constexpr int COLS = BLOCK_N / 16;
+        const int TILE_SIZE = TILE_N * sizeof(block_q8_0) + TILE_N * sizeof(int32_t);
+
+        const block_q8_0 * RESTRICT A = static_cast(_A);
+        const char * RESTRICT B = static_cast(_B);
+
+        __m512i va[8];
+        __m512i vb[8];
+        __m512 vc[COLS];
+        __m512 vd1;
+
+        // Notes: s8s8 igemm compensation in avx512-vnni
+        // change s8s8 to u8s8 with compensate
+        //   a * b = (a + 128) * b - 128 * b
+        //   s   s       u       s    u    s
+        //
+        // (128 * b is pre-computed when packing B to vnni formats)
+        //
+        const __m512i off = _mm512_set1_epi8(static_cast(0x80));
+
+        auto loadc = [&](int col) {
+            vc[col] = _mm512_setzero_ps();
+        };
+        Unroll{}(loadc);
+
+        auto compute = [&](int col, int i) {
+            // load a and add offset 128
+            if (col == 0) {
+                const int32_t * a_ptr = reinterpret_cast(A[0 * KB + i].qs);
+                for (int k = 0; k < 8; ++k) {
+                    va[k] = _mm512_set1_epi32(a_ptr[k]);
+                    va[k] = _mm512_add_epi8(va[k], off);
+                }
+                vd1 = _mm512_set1_ps(GGML_FP16_TO_FP32(A[0 * KB + i].d));
+            }
+
+            // load b
+            const char * b_ptr = B + PACKED_INDEX(col, i, KB, TILE_SIZE);
+            for (int k = 0; k < 8; ++k) {
+                vb[k] = _mm512_loadu_si512((const __m512i *)(b_ptr + k * 64));
+            }
+            const int offset = TILE_N * TILE_K;
+            const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(b_ptr + offset)));
+            const int offset2 = TILE_N * TILE_K + TILE_N * sizeof(ggml_half);
+            const __m512i vcomp = _mm512_loadu_si512((const __m512i *)(b_ptr + offset2));
+
+            __m512i vsum = _mm512_setzero_si512();
+            for (int k = 0; k < 8; ++k) {
+                vsum = _mm512_dpbusd_epi32(vsum, va[k], vb[k]);
+            }
+            vsum = _mm512_sub_epi32(vsum, vcomp);
+
+            vc[col] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(vsum), _mm512_mul_ps(vd0, vd1), vc[col]);
+        };
+
+        for (int i = 0; i < KB; ++i) {
+            Unroll{}(compute, i);
+        }
+
+        //store to C
+        auto storec = [&](int col) {
+            _mm512_storeu_ps((__m512i*)(C + 0 * ldc + col * 16), vc[col]);
+        };
+        Unroll{}(storec);
+    }
+};
+
+template 
+struct tinygemm_kernel_vnni {
+    static void apply(int KB, const void * RESTRICT _A, const void * RESTRICT _B, float * RESTRICT C, int ldc) {
+
+        constexpr int COLS = BLOCK_N / 16;
+        const int TILE_SIZE = TILE_N * sizeof(block_q4_K) + TILE_N * 4;
+
+        const block_q8_K * RESTRICT A = static_cast(_A);
+        const char * RESTRICT B = static_cast(_B);
+
+        // a.qs:   8 groups, 32 bytes each group (m256i)
+        __m512i va[8];
+        // a.bsum: 8 groups,  2 bytes each group (m128i)
+        __m512i va_bsum;
+        __m512 vc[COLS];
+        __m512 vd1;
+
+        // packed_B:
+        const int offset_scales = (QK_K / 2) * TILE_N;
+        const int offset_mins   = (QK_K / 2) * TILE_N +  8 * TILE_N;
+        const int offset_d0     = (QK_K / 2) * TILE_N + 16 * TILE_N;
+        const int offset_dmin   = (QK_K / 2) * TILE_N + 16 * TILE_N + TILE_N * sizeof(ggml_half);
+
+        const __m512i lowMask = _mm512_set1_epi8(0xF);
+
+        auto loadc = [&](int col) {
+            vc[col] = _mm512_setzero_ps();
+        };
+        Unroll{}(loadc);
+
+        // Notes: vnni formats in QK_K
+        //   a) quants vnni format
+        //     int8  {k/4, n, 4}, viewed as 2d {k/4, 4n}, k = 32
+        //     from {16, 32} to {8, 64}
+        //
+        //   b) min vnni format
+        //     int16 {k/2, n, 2}, viewed as 2d {k/2, 2n}, k = 8
+        //     from {16,  8} to {4, 32}
+        //
+        auto compute = [&](int col, int i) {
+            // load a
+            if (col == 0) {
+                for (int k_group = 0; k_group < QK_K / 32; ++k_group) {
+                    va[k_group] = _mm512_castsi256_si512(_mm256_loadu_si256((const __m256i *)(A[0 * KB + i].qs + k_group * 32)));
+                }
+                const __m256i q8sums = _mm256_loadu_si256((const __m256i *)A[0 * KB + i].bsums);
+                const __m128i q8s = _mm_hadd_epi16(_mm256_extracti128_si256(q8sums, 0), _mm256_extracti128_si256(q8sums, 1));
+                va_bsum = _mm512_castsi128_si512(q8s);
+                vd1 = _mm512_set1_ps(A[0 * KB + i].d);
+            }
+
+            // step 1: accumultate the quants
+            __m512i acc = _mm512_setzero_si512();
+            const char * b_ptr = B + PACKED_INDEX(col, i, KB, TILE_SIZE);
+            const char * b_qs  = b_ptr;
+            for (int k_group = 0; k_group < QK_K / 32; ++k_group) {
+                __m512i vsum = _mm512_setzero_si512();
+                for (int k = 0; k < 8; k += 2) {
+                    __m512i va0 = _mm512_permutexvar_epi32(_mm512_set1_epi32(k + 0), va[k_group]);
+                    __m512i va1 = _mm512_permutexvar_epi32(_mm512_set1_epi32(k + 1), va[k_group]);
+
+                    __m512i bytes = _mm512_loadu_si512((const __m512i *)b_qs);
+                    __m512i vb0 = _mm512_and_si512(bytes, lowMask);
+                    vsum = _mm512_dpbusd_epi32(vsum, vb0, va0);
+                    __m512i vb1 = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+                    vsum = _mm512_dpbusd_epi32(vsum, vb1, va1);
+
+                    b_qs += 64;
+                }
+                // vacc += scale * (q8 @ q4)
+                const __m512i vscale = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i *)(b_ptr + offset_scales + k_group * TILE_N)));
+                acc = _mm512_add_epi32(acc, _mm512_mullo_epi32(vsum, vscale));
+            }
+            const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(b_ptr + offset_d0)));
+            vc[col] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(acc), _mm512_mul_ps(vd0, vd1), vc[col]);
+
+            // step 2: accumulate the mins
+            __m512i acc_m = _mm512_setzero_si512();
+            for (int k = 0; k < 4; ++k) {
+                __m512i vmask = _mm512_set1_epi32(k);
+                __m512i va = _mm512_permutexvar_epi32(vmask, va_bsum);
+                __m512i vb = _mm512_cvtepi8_epi16(_mm256_loadu_si256((const __m256i *)(b_ptr + offset_mins + k * 32)));
+                acc_m = _mm512_dpwssds_epi32(acc_m, va, vb);
+            }
+            const __m512 vdmin = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(b_ptr + offset_dmin)));
+            vc[col] = _mm512_fnmadd_ps(_mm512_cvtepi32_ps(acc_m), _mm512_mul_ps(vdmin, vd1), vc[col]);
+        };
+
+        for (int i = 0; i < KB; ++i) {
+            Unroll{}(compute, i);
+        }
+
+        //store to C
+        auto storec = [&](int col) {
+            _mm512_storeu_ps((__m512i*)(C + 0 * ldc + col * 16), vc[col]);
+        };
+        Unroll{}(storec);
+    }
+};
+
+template 
+struct tinygemm_kernel_vnni {
+    static void apply(int KB, const void * RESTRICT _A, const void * RESTRICT _B, float * RESTRICT C, int ldc) {
+
+        constexpr int COLS = BLOCK_N / 16;
+        const int TILE_SIZE = TILE_N * sizeof(block_q5_K) + TILE_N * 4;
+
+        const block_q8_K * RESTRICT A = static_cast(_A);
+        const char * RESTRICT B = static_cast(_B);
+
+        // a.qs:   8 groups, 32 bytes each group (m256i)
+        __m512i va[8];
+        // a.bsum: 8 groups,  2 bytes each group (m128i)
+        __m512i va_bsum;
+        __m512 vc[COLS];
+        __m512 vd1;
+
+        // packed_B:
+        const int offset_qh     = (QK_K / 2) * TILE_N;
+        const int offset_scales = (QK_K / 2) * TILE_N + (QK_K / 8) * TILE_N;
+        const int offset_mins   = (QK_K / 2) * TILE_N + (QK_K / 8) * TILE_N +  8 * TILE_N;
+        const int offset_d0     = (QK_K / 2) * TILE_N + (QK_K / 8) * TILE_N + 16 * TILE_N;
+        const int offset_dmin   = (QK_K / 2) * TILE_N + (QK_K / 8) * TILE_N + 16 * TILE_N + TILE_N * sizeof(ggml_half);
+
+        const __m512i lowMask = _mm512_set1_epi8(0xF);
+
+        auto loadc = [&](int col) {
+            vc[col] = _mm512_setzero_ps();
+        };
+        Unroll{}(loadc);
+
+        // Q5_K and Q4_K shares the same vnni formats, refer to notes above.
+        auto compute = [&](int col, int i) {
+            // load a
+            if (col == 0) {
+                for (int k_group = 0; k_group < QK_K / 32; ++k_group) {
+                    va[k_group] = _mm512_castsi256_si512(_mm256_loadu_si256((const __m256i *)(A[0 * KB + i].qs + k_group * 32)));
+                }
+                const __m256i q8sums = _mm256_loadu_si256((const __m256i *)A[0 * KB + i].bsums);
+                const __m128i q8s = _mm_hadd_epi16(_mm256_extracti128_si256(q8sums, 0), _mm256_extracti128_si256(q8sums, 1));
+                va_bsum = _mm512_castsi128_si512(q8s);
+                vd1 = _mm512_set1_ps(A[0 * KB + i].d);
+            }
+
+            // step 1: accumultate the quants
+            __m512i acc = _mm512_setzero_si512();
+            const char * b_ptr = B + PACKED_INDEX(col, i, KB, TILE_SIZE);
+            const char * b_qs  = b_ptr;
+            const char * b_qh  = b_ptr + offset_qh;
+            for (int k_group = 0; k_group < QK_K / 32; ++k_group) {
+                __m512i vsum = _mm512_setzero_si512();
+                __m512i hmask0 = _mm512_set1_epi8(0x1);
+                __m512i hmask1 = _mm512_set1_epi8(0x2);
+                __m512i hbits = _mm512_loadu_si512((const __m512i *)(b_qh + k_group * 64));
+                for (int k = 0; k < 8; k += 2) {
+                    __m512i va0 = _mm512_permutexvar_epi32(_mm512_set1_epi32(k + 0), va[k_group]);
+                    __m512i va1 = _mm512_permutexvar_epi32(_mm512_set1_epi32(k + 1), va[k_group]);
+
+                    __m512i bytes = _mm512_loadu_si512((const __m512i *)b_qs);
+                    __m512i vb0 = _mm512_and_si512(bytes, lowMask);
+                    __m512i vb1 = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+
+                    __m512i vh0 = _mm512_slli_epi16(_mm512_srli_epi16(_mm512_and_si512(hbits, hmask0), k), 4);
+                    __m512i vh1 = _mm512_slli_epi16(_mm512_srli_epi16(_mm512_and_si512(hbits, hmask1), k + 1), 4);
+
+                    hmask0 = _mm512_slli_epi16(hmask0, 2);
+                    hmask1 = _mm512_slli_epi16(hmask1, 2);
+                    vb0 = _mm512_add_epi8(vb0, vh0);
+                    vb1 = _mm512_add_epi8(vb1, vh1);
+
+                    vsum = _mm512_dpbusd_epi32(vsum, vb0, va0);
+                    vsum = _mm512_dpbusd_epi32(vsum, vb1, va1);
+
+                    b_qs += 64;
+                }
+                // vacc += scale * (q8 @ q5)
+                const __m512i vscale = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i *)(b_ptr + offset_scales + k_group * TILE_N)));
+                acc = _mm512_add_epi32(acc, _mm512_mullo_epi32(vsum, vscale));
+            }
+            const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(b_ptr + offset_d0)));
+            vc[col] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(acc), _mm512_mul_ps(vd0, vd1), vc[col]);
+
+            // step 2: accumulate the mins
+            __m512i acc_m = _mm512_setzero_si512();
+            for (int k = 0; k < 4; ++k) {
+                __m512i vmask = _mm512_set1_epi32(k);
+                __m512i va = _mm512_permutexvar_epi32(vmask, va_bsum);
+                __m512i vb = _mm512_cvtepi8_epi16(_mm256_loadu_si256((const __m256i *)(b_ptr + offset_mins + k * 32)));
+                acc_m = _mm512_dpwssds_epi32(acc_m, va, vb);
+            }
+            const __m512 vdmin = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(b_ptr + offset_dmin)));
+            vc[col] = _mm512_fnmadd_ps(_mm512_cvtepi32_ps(acc_m), _mm512_mul_ps(vdmin, vd1), vc[col]);
+        };
+
+        for (int i = 0; i < KB; ++i) {
+            Unroll{}(compute, i);
+        }
+
+        //store to C
+        auto storec = [&](int col) {
+            _mm512_storeu_ps((__m512i*)(C + 0 * ldc + col * 16), vc[col]);
+        };
+        Unroll{}(storec);
+    }
+};
+
+template 
+struct tinygemm_kernel_vnni {
+    static void apply(int KB, const void * RESTRICT _A, const void * RESTRICT _B, float * RESTRICT C, int ldc) {
+
+        constexpr int COLS = BLOCK_N / 16;
+        const int TILE_SIZE = TILE_N * sizeof(block_q6_K);
+
+        const block_q8_K * RESTRICT A = static_cast(_A);
+        const char * RESTRICT B = static_cast(_B);
+
+        // load the 256 bytes from A to 4 avx512 vectors
+        __m512i va[4];
+        __m512 vc[COLS];
+        __m512 vd1;
+
+        // packed_B:
+        const int offset_qh     = (QK_K / 2) * TILE_N;
+        const int offset_scales = (QK_K / 2) * TILE_N + (QK_K / 4) * TILE_N;
+        const int offset_d0     = (QK_K / 2) * TILE_N + (QK_K / 4) * TILE_N + 16 * TILE_N;
+
+        // compensation
+        __m512i vcomp;
+
+        const __m512i m32s = _mm512_set1_epi32(32);
+        const __m512i lowMask = _mm512_set1_epi8(0xF);
+
+        auto loadc = [&](int col) {
+            vc[col] = _mm512_setzero_ps();
+        };
+        Unroll{}(loadc);
+
+        auto compute = [&](int col, int i) {
+            if (col == 0) {
+                // load a
+                va[0] = _mm512_loadu_si512((const __m512i *)(A[0 * KB + i].qs +   0));
+                va[1] = _mm512_loadu_si512((const __m512i *)(A[0 * KB + i].qs +  64));
+                va[2] = _mm512_loadu_si512((const __m512i *)(A[0 * KB + i].qs + 128));
+                va[3] = _mm512_loadu_si512((const __m512i *)(A[0 * KB + i].qs + 192));
+
+                const __m256i q8sums = _mm256_loadu_si256((const __m256i *)A[0 * KB + i].bsums);
+                vcomp = _mm512_mullo_epi32(_mm512_cvtepi16_epi32(q8sums), m32s);
+                vd1 = _mm512_set1_ps(A[0 * KB + i].d);
+            }
+
+            // accmulate the quants
+            __m512i acc = _mm512_setzero_si512();
+            const char * b_ptr = B + PACKED_INDEX(col, i, KB, TILE_SIZE);
+            const char * b_qs = b_ptr;
+            const char * b_qh = b_ptr + offset_qh;
+            int mask = 0;
+            for (int k_group = 0; k_group < QK_K / 16; ++k_group) {
+                int r = k_group >> 2;
+                __m512i va0 = _mm512_permutexvar_epi32(_mm512_set1_epi32(mask++), va[r]);
+                __m512i va1 = _mm512_permutexvar_epi32(_mm512_set1_epi32(mask++), va[r]);
+
+                __m512i vsum = _mm512_setzero_si512();
+                __m512i hmask = _mm512_set1_epi8(0x3);
+
+                __m512i bytes = _mm512_loadu_si512(b_qs);
+                __m512i hbits = _mm512_loadu_si512(b_qh);
+                __m512i vb0 = _mm512_and_si512(bytes, lowMask);
+                __m512i vb1 = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+                __m512i vh0 = _mm512_slli_epi16(_mm512_and_si512(hbits, hmask), 4);
+                __m512i vh1 = _mm512_slli_epi16(_mm512_and_si512(hbits, _mm512_slli_epi16(hmask, 2)), 2);
+
+                vb0 = _mm512_add_epi8(vb0, vh0);
+                vb1 = _mm512_add_epi8(vb1, vh1);
+                vsum = _mm512_dpbusd_epi32(vsum, vb0, va0);
+                vsum = _mm512_dpbusd_epi32(vsum, vb1, va1);
+                b_qs += 64;
+
+                va0 = _mm512_permutexvar_epi32(_mm512_set1_epi32(mask++), va[r]);
+                va1 = _mm512_permutexvar_epi32(_mm512_set1_epi32(mask++), va[r]);
+
+                bytes = _mm512_loadu_si512(b_qs);
+                vb0 = _mm512_and_si512(bytes, lowMask);
+                vb1 = _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask);
+                vh0 =                   _mm512_and_si512(hbits, _mm512_slli_epi16(hmask, 4));
+                vh1 = _mm512_srli_epi16(_mm512_and_si512(hbits, _mm512_slli_epi16(hmask, 6)), 2);
+                vb0 = _mm512_add_epi8(vb0, vh0);
+                vb1 = _mm512_add_epi8(vb1, vh1);
+                vsum = _mm512_dpbusd_epi32(vsum, vb0, va0);
+                vsum = _mm512_dpbusd_epi32(vsum, vb1, va1);
+                b_qs += 64;
+                b_qh += 64;
+
+                // B * A - 32 * A
+                __m512i vmask = _mm512_set1_epi32(k_group);
+                vsum = _mm512_sub_epi32(vsum, _mm512_permutexvar_epi32(vmask, vcomp));
+
+                // vacc += scale * (q8 @ q6)
+                const __m512i vscale = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i *)(b_ptr + offset_scales + k_group * TILE_N)));
+                acc = _mm512_add_epi32(acc, _mm512_mullo_epi32(vsum, vscale));
+            }
+            const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(b_ptr + offset_d0)));
+            vc[col] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(acc), _mm512_mul_ps(vd0, vd1), vc[col]);
+        };
+
+        for (int i = 0; i < KB; ++i) {
+            Unroll{}(compute, i);
+        }
+
+        //store to C
+        auto storec = [&](int col) {
+            _mm512_storeu_ps((__m512i*)(C + 0 * ldc + col * 16), vc[col]);
+        };
+        Unroll{}(storec);
+    }
+};
+
+template 
+struct tinygemm_kernel_vnni {
+    static void apply(int KB, const void * RESTRICT _A, const void * RESTRICT _B, float * RESTRICT C, int ldc) {
+
+        constexpr int COLS = BLOCK_N / 16;
+        const int TILE_SIZE = TILE_N * sizeof(block_iq4_xs) + TILE_N * 2;
+
+        const block_q8_K * RESTRICT A = static_cast(_A);
+        const char * RESTRICT B = static_cast(_B);
+
+        // load the 256 bytes from A to 4 avx512 vectors
+        __m512i va[4];
+        __m512 vc[COLS];
+        __m512 vd1;
+
+        // packed_B:
+        const int offset_scales = (QK_K / 2) * TILE_N ;
+        const int offset_d0     = (QK_K / 2) * TILE_N + 8 * TILE_N;
+
+        // compensation
+        __m512i vcomp;
+
+        const __m256i m128s = _mm256_set1_epi16(128);
+        const __m512i lowMask = _mm512_set1_epi8(0xF);
+
+        const __m512i values128 = _mm512_set_epi8(
+            113, 89, 69, 53, 38, 25, 13, 1, -10, -22, -35, -49, -65, -83, -104, -127,
+            113, 89, 69, 53, 38, 25, 13, 1, -10, -22, -35, -49, -65, -83, -104, -127,
+            113, 89, 69, 53, 38, 25, 13, 1, -10, -22, -35, -49, -65, -83, -104, -127,
+            113, 89, 69, 53, 38, 25, 13, 1, -10, -22, -35, -49, -65, -83, -104, -127
+        );
+        const __m512i off = _mm512_set1_epi8(static_cast(0x80));
+        const __m512i values256 = _mm512_add_epi8(values128, off);
+
+        auto loadc = [&](int col) {
+            vc[col] = _mm512_setzero_ps();
+        };
+        Unroll{}(loadc);
+
+        auto compute = [&](int col, int i) {
+            if (col == 0) {
+                // load a
+                va[0] = _mm512_loadu_si512((const __m512i *)(A[0 * KB + i].qs +   0));
+                va[1] = _mm512_loadu_si512((const __m512i *)(A[0 * KB + i].qs +  64));
+                va[2] = _mm512_loadu_si512((const __m512i *)(A[0 * KB + i].qs + 128));
+                va[3] = _mm512_loadu_si512((const __m512i *)(A[0 * KB + i].qs + 192));
+
+                // compensation: 128 * A
+                const __m256i q8sums = _mm256_loadu_si256((const __m256i *)A[0 * KB + i].bsums);
+                vcomp = _mm512_castsi256_si512(_mm256_madd_epi16(q8sums, m128s));
+                vd1 = _mm512_set1_ps(A[0 * KB + i].d);
+            }
+
+            // accmulate the quants
+            __m512i acc = _mm512_setzero_si512();
+            const char * b_ptr = B + PACKED_INDEX(col, i, KB, TILE_SIZE);
+            const char * b_qs = b_ptr;
+            int mask = 0;
+            for (int k_group = 0; k_group < QK_K / 32; ++k_group) {
+                int r = k_group >> 1;
+                __m512i vmask = _mm512_set1_epi32(k_group);
+                __m512i vsum = _mm512_setzero_si512();
+                for (int k = 0; k < 8; k += 2) {
+                    __m512i va0 = _mm512_permutexvar_epi32(_mm512_set1_epi32(mask++), va[r]);
+                    __m512i va1 = _mm512_permutexvar_epi32(_mm512_set1_epi32(mask++), va[r]);
+
+                    __m512i bytes = _mm512_loadu_si512(b_qs);
+                    __m512i vb0 = _mm512_shuffle_epi8(values256, _mm512_and_si512(bytes, lowMask));
+                    __m512i vb1 = _mm512_shuffle_epi8(values256, _mm512_and_si512(_mm512_srli_epi16(bytes, 4), lowMask));
+
+                    vsum = _mm512_dpbusd_epi32(vsum, vb0, va0);
+                    vsum = _mm512_dpbusd_epi32(vsum, vb1, va1);
+                    b_qs += 64;
+                }
+                // (B + 128) * A - 128 * A
+                vsum = _mm512_sub_epi32(vsum, _mm512_permutexvar_epi32(vmask, vcomp));
+
+                // vacc += scale * (q8 @ q4)
+                const __m512i vscale = _mm512_cvtepi8_epi32(_mm_loadu_si128((const __m128i *)(b_ptr + offset_scales + k_group * TILE_N)));
+                acc = _mm512_add_epi32(acc, _mm512_mullo_epi32(vsum, vscale));
+            }
+            const __m512 vd0 = _mm512_cvtph_ps(_mm256_loadu_si256((const __m256i *)(b_ptr + offset_d0)));
+            vc[col] = _mm512_fmadd_ps(_mm512_cvtepi32_ps(acc), _mm512_mul_ps(vd0, vd1), vc[col]);
+        };
+
+        for (int i = 0; i < KB; ++i) {
+            Unroll{}(compute, i);
+        }
+
+        //store to C
+        auto storec = [&](int col) {
+            _mm512_storeu_ps((__m512i*)(C + 0 * ldc + col * 16), vc[col]);
+        };
+        Unroll{}(storec);
+    }
+};
+
+#define LAUNCH_TINYGEMM_KERNEL_VNNI(NB_SIZE)                                         \
+    tinygemm_kernel_vnni::apply(   \
+        KB, (const char *)wdata + 0 * row_size_A,                                    \
+        (const char *)src0->data + PACKED_INDEX(nb * kTilesN, 0, KB, TILE_SIZE),     \
+        (float *) dst->data + 0 * N + nb_start, ldc)
+
+template ::value, int>::type = 0>
+void tinygemm_kernel_amx(int M, int N, int KB, const void * RESTRICT _A, const void * RESTRICT _B, TC * RESTRICT C, int ldc) {
+    using packed_B_t = packed_B_type;
+    const int TILE_SIZE = get_tile_size();
+    const bool need_unpack = do_unpack::value;
+
+    GGML_ASSERT(M <= 2 * TILE_M && N == 2 * TILE_N);
+    const TA * RESTRICT A = static_cast(_A);
+    const char * RESTRICT B = static_cast(_B);
+
+    const int m0 = std::min(M, TILE_M);
+    const int m1 = std::max(M - TILE_M, 0);
+    const int lda = KB * sizeof(TA);
+    //const int ldb = KB * sizeof(TB);
+
+    static thread_local packed_B_t Tile0[TILE_N * TILE_K];
+    static thread_local packed_B_t Tile1[TILE_N * TILE_K];
+    static thread_local int8_t Tile23[TILE_M * TILE_K];
+
+    static thread_local int32_t TileC0[TILE_M * TILE_N * 4];
+    static thread_local int32_t TileC1[TILE_M * TILE_N * 4];
+
+    // double buffering C to interleave avx512 and amx
+    int32_t * C_cur = TileC0;
+    int32_t * C_pre = TileC1;
+
+    auto Tile4 = [&](int32_t * base) { return base; };
+    auto Tile5 = [&](int32_t * base) { return base + TILE_M * TILE_N; };
+    auto Tile6 = [&](int32_t * base) { return base + 2 * TILE_M * TILE_N; };
+    auto Tile7 = [&](int32_t * base) { return base + 3 * TILE_M * TILE_N; };
+
+    if (M == 2 * TILE_M) {
+        // i = 0
+        const char * B_blk0 = B + PACKED_INDEX(0, 0, KB, TILE_SIZE);
+        const char * B_blk1 = B + PACKED_INDEX(1, 0, KB, TILE_SIZE);
+        if (need_unpack) {
+            unpack_B(Tile0, B_blk0);
+            _tile_loadd(TMM0, Tile0, TILE_N * VNNI_BLK);
+        } else {
+            _tile_loadd(TMM0, B_blk0, TILE_N * VNNI_BLK);
+        }
+
+        _tile_zero(TMM4);
+        _tile_loadd(TMM2, A[0].qs, lda);
+        _tile_dpbssd(TMM4, TMM2, TMM0);
+        _tile_stored(TMM4, Tile4(C_pre), TILE_N * sizeof(int32_t));
+
+        _tile_zero(TMM5);
+        _tile_loadd(TMM3, A[TILE_M * KB + 0].qs, lda);
+        _tile_dpbssd(TMM5, TMM3, TMM0);
+        _tile_stored(TMM5, Tile5(C_pre), TILE_N * sizeof(int32_t));
+
+        if (need_unpack) {
+            unpack_B(Tile1, B_blk0);
+            _tile_loadd(TMM1, Tile1, TILE_N * VNNI_BLK);
+        } else {
+            _tile_loadd(TMM1, B_blk1, TILE_N * VNNI_BLK);
+        }
+
+        _tile_zero(TMM6);
+        _tile_dpbssd(TMM6, TMM2, TMM1);
+        _tile_stored(TMM6, Tile6(C_pre), TILE_N * sizeof(int32_t));
+
+        _tile_zero(TMM7);
+        _tile_dpbssd(TMM7, TMM3, TMM1);
+        _tile_stored(TMM7, Tile7(C_pre), TILE_N * sizeof(int32_t));
+
+        for (int i = 1; i < KB; ++i) {
+            // index of previous iter
+            const int ii = i - 1;
+            const char * B_blk0 = B + PACKED_INDEX(0, i, KB, TILE_SIZE);
+            const char * B_blk1 = B + PACKED_INDEX(1, i, KB, TILE_SIZE);
+            GGML_DISPATCH_BOOL(ii > 0, is_acc, [&] {
+                if (need_unpack) {
+                    unpack_B(Tile0, B_blk0);
+                    _tile_loadd(TMM0, Tile0, TILE_N * VNNI_BLK);
+                } else {
+                    _tile_loadd(TMM0, B_blk0, TILE_N * VNNI_BLK);
+                }
+                _tile_zero(TMM4);
+                _tile_loadd(TMM2, A[i].qs, lda);
+                acc_C::apply(C, ldc, Tile4(C_pre), &A[ii], KB, B + PACKED_INDEX(0, ii, KB, TILE_SIZE), TILE_M);
+
+                _tile_dpbssd(TMM4, TMM2, TMM0);
+                _tile_stored(TMM4, Tile4(C_cur), TILE_N * sizeof(int32_t));
+
+                _tile_zero(TMM5);
+                _tile_loadd(TMM3, A[TILE_M * KB + i].qs, lda);
+                acc_C::apply(C + TILE_M * ldc, ldc, Tile5(C_pre), &A[TILE_M * KB + ii], KB, B + PACKED_INDEX(0, ii, KB, TILE_SIZE), TILE_M);
+
+                _tile_dpbssd(TMM5, TMM3, TMM0);
+                _tile_stored(TMM5, Tile5(C_cur), TILE_N * sizeof(int32_t));
+
+                if (need_unpack) {
+                    unpack_B(Tile1, B_blk1);
+                    _tile_loadd(TMM1, Tile1, TILE_N * VNNI_BLK);
+                } else {
+                    _tile_loadd(TMM1, B_blk1, TILE_N * VNNI_BLK);
+                }
+                _tile_zero(TMM6);
+                acc_C::apply(C + TILE_N, ldc, Tile6(C_pre), &A[ii], KB, B + PACKED_INDEX(1, ii, KB, TILE_SIZE), TILE_M);
+
+                _tile_dpbssd(TMM6, TMM2, TMM1);
+                _tile_stored(TMM6, Tile6(C_cur), TILE_N * sizeof(int32_t));
+
+                _tile_zero(TMM7);
+                acc_C::apply(C + TILE_M * ldc + TILE_N, ldc, Tile7(C_pre), &A[TILE_M * KB + ii], KB, B + PACKED_INDEX(1, ii, KB, TILE_SIZE), TILE_M);
+
+                _tile_dpbssd(TMM7, TMM3, TMM1);
+                _tile_stored(TMM7, Tile7(C_cur), TILE_N * sizeof(int32_t));
+
+                std::swap(C_cur, C_pre);
+            });
+        }
+        // final accumulation
+        {
+            int ii = KB - 1;
+            acc_C::apply(C, ldc, Tile4(C_pre), &A[ii], KB, B + PACKED_INDEX(0, ii, KB, TILE_SIZE), TILE_M);
+            acc_C::apply(C + TILE_M * ldc, ldc, Tile5(C_pre), &A[TILE_M * KB + ii], KB, B + PACKED_INDEX(0, ii, KB, TILE_SIZE), TILE_M);
+            acc_C::apply(C + TILE_N, ldc, Tile6(C_pre), &A[ii], KB, B + PACKED_INDEX(1, ii, KB, TILE_SIZE), TILE_M);
+            acc_C::apply(C + TILE_M * ldc + TILE_N, ldc, Tile7(C_pre), &A[TILE_M * KB + ii], KB, B + PACKED_INDEX(1, ii, KB, TILE_SIZE), TILE_M);
+        }
+    } else {
+        for (int i = 0; i < KB; ++i) {
+            _tile_zero(TMM4);
+            _tile_zero(TMM6);
+            if (m1 != 0) {
+                _tile_zero(TMM5);
+                _tile_zero(TMM7);
+            }
+
+            const char * B_blk0 = B + PACKED_INDEX(0, i, KB, TILE_SIZE);
+            const char * B_blk1 = B + PACKED_INDEX(1, i, KB, TILE_SIZE);
+            if (need_unpack) {
+                unpack_B(Tile0, B_blk0);
+                _tile_loadd(TMM0, Tile0, TILE_N * VNNI_BLK);
+            } else {
+                _tile_loadd(TMM0, B_blk0, TILE_N * VNNI_BLK);
+            }
+
+            if (need_unpack) {
+                unpack_B(Tile1, B_blk1);
+                _tile_loadd(TMM1, Tile1, TILE_N * VNNI_BLK);
+            } else {
+                _tile_loadd(TMM1, B_blk1, TILE_N * VNNI_BLK);
+            }
+
+            if (m0 == TILE_M) {
+                _tile_loadd(TMM2, A[i].qs, lda);
+            } else {
+                unpack_A(Tile23, &A[i], KB, m0);
+                _tile_loadd(TMM2, Tile23, TILE_K);
+            }
+
+            _tile_dpbssd(TMM4, TMM2, TMM0);
+            _tile_dpbssd(TMM6, TMM2, TMM1);
+
+            _tile_stored(TMM4, Tile4(C_cur), TILE_N * sizeof(int32_t));
+            _tile_stored(TMM6, Tile6(C_cur), TILE_N * sizeof(int32_t));
+
+            GGML_DISPATCH_BOOL(i > 0, is_acc, [&] {
+                acc_C::apply(C,          ldc, Tile4(C_cur), &A[i], KB, B + PACKED_INDEX(0, i, KB, TILE_SIZE), m0);
+                acc_C::apply(C + TILE_N, ldc, Tile6(C_cur), &A[i], KB, B + PACKED_INDEX(1, i, KB, TILE_SIZE), m0);
+            });
+
+            if (m1 != 0) {
+                unpack_A(Tile23, &A[TILE_M * KB + i], KB, m1);
+                _tile_loadd(TMM3, Tile23, TILE_K);
+
+                _tile_dpbssd(TMM5, TMM3, TMM0);
+                _tile_dpbssd(TMM7, TMM3, TMM1);
+                _tile_stored(TMM5, Tile5(C_cur), TILE_N * sizeof(int32_t));
+                _tile_stored(TMM7, Tile7(C_cur), TILE_N * sizeof(int32_t));
+                GGML_DISPATCH_BOOL(i > 0, is_acc, [&] {
+                    acc_C::apply(C + TILE_M * ldc,          ldc, Tile5(C_cur), &A[TILE_M * KB + i], KB, B + PACKED_INDEX(0, i, KB, TILE_SIZE), m1);
+                    acc_C::apply(C + TILE_M * ldc + TILE_N, ldc, Tile7(C_cur), &A[TILE_M * KB + i], KB, B + PACKED_INDEX(1, i, KB, TILE_SIZE), m1);
+                });
+            }
+        }
+    }
+    return;
+}
+
+template ::value, int>::type = 0>
+void tinygemm_kernel_amx(int M, int N, int KB, const void * RESTRICT _A, const void * RESTRICT _B, float * RESTRICT C, int ldc) {
+    static_assert(std::is_same::value);
+    const int TILE_SIZE = get_tile_size();
+
+    GGML_ASSERT(M <= 2 * TILE_M && N == 2 * TILE_N);
+    const TA * RESTRICT A = static_cast(_A);
+    const char * RESTRICT B = static_cast(_B);
+
+    const int m0 = std::min(M, TILE_M);
+    const int m1 = std::max(M - TILE_M, 0);
+    //const int lda = KB * sizeof(TA);
+
+    static thread_local int8_t Tile0[TILE_N * TILE_K];
+    static thread_local int8_t Tile1[TILE_N * TILE_K];
+    static thread_local int8_t Tile23[TILE_M * TILE_K];
+
+    // mat mul result for each group
+    static thread_local int32_t Tile4[TILE_M * TILE_N];
+    static thread_local int32_t Tile5[TILE_M * TILE_N];
+    static thread_local int32_t Tile6[TILE_M * TILE_N];
+    static thread_local int32_t Tile7[TILE_M * TILE_N];
+
+    // sum of each QK_K block, contains 8 groups, int32
+    static thread_local int32_t Sumi4[TILE_M * TILE_N];
+    static thread_local int32_t Sumi5[TILE_M * TILE_N];
+    static thread_local int32_t Sumi6[TILE_M * TILE_N];
+    static thread_local int32_t Sumi7[TILE_M * TILE_N];
+
+    const int k_group_size = std::is_same::value ? 16 : 32;
+    for (int i = 0; i < KB; ++i) {
+        // step 1: accumulate the quants across 8 groups, each group with 32
+        for (int k = 0; k < QK_K / k_group_size; ++k) {
+            GGML_DISPATCH_BOOL(k > 0, is_acc, [&] {
+                _tile_zero(TMM4);
+                _tile_zero(TMM6);
+
+                unpack_B(Tile0, B + PACKED_INDEX(0, i, KB, TILE_SIZE), k);
+                _tile_loadd(TMM0, Tile0, TILE_N * VNNI_BLK);
+
+                unpack_B(Tile1, B + PACKED_INDEX(1, i, KB, TILE_SIZE), k);
+                _tile_loadd(TMM1, Tile1, TILE_N * VNNI_BLK);
+
+                unpack_A(Tile23, &A[i], KB, k, m0);
+                _tile_loadd(TMM2, Tile23, TILE_K);
+
+                _tile_dpbssd(TMM4, TMM2, TMM0);
+                _tile_dpbssd(TMM6, TMM2, TMM1);
+
+                _tile_stored(TMM4, Tile4, TILE_N * sizeof(int32_t));
+                _tile_stored(TMM6, Tile6, TILE_N * sizeof(int32_t));
+
+                scale_C(Tile4, Sumi4, B + PACKED_INDEX(0, i, KB, TILE_SIZE), k, m0);
+                scale_C(Tile6, Sumi6, B + PACKED_INDEX(1, i, KB, TILE_SIZE), k, m0);
+
+                if (m1 != 0) {
+                    _tile_zero(TMM5);
+                    _tile_zero(TMM7);
+
+                    unpack_A(Tile23, &A[TILE_M * KB + i], KB, k, m1);
+                    _tile_loadd(TMM3, Tile23, TILE_K);
+
+                    _tile_dpbssd(TMM5, TMM3, TMM0);
+                    _tile_dpbssd(TMM7, TMM3, TMM1);
+
+                    _tile_stored(TMM5, Tile5, TILE_N * sizeof(int32_t));
+                    _tile_stored(TMM7, Tile7, TILE_N * sizeof(int32_t));
+
+                    scale_C(Tile5, Sumi5, B + PACKED_INDEX(0, i, KB, TILE_SIZE), k, m1);
+                    scale_C(Tile7, Sumi7, B + PACKED_INDEX(1, i, KB, TILE_SIZE), k, m1);
+                }
+            });
+        }
+
+        // step 2: accmulate the mins
+        GGML_DISPATCH_BOOL(i > 0, is_acc, [&] {
+            acc_C::apply(C,          ldc, Sumi4, &A[i], KB, B + PACKED_INDEX(0, i, KB, TILE_SIZE), m0);
+            acc_C::apply(C + TILE_N, ldc, Sumi6, &A[i], KB, B + PACKED_INDEX(1, i, KB, TILE_SIZE), m0);
+            if (m1 != 0) {
+                acc_C::apply(C + TILE_M * ldc,          ldc, Sumi5, &A[TILE_M * KB + i], KB, B + PACKED_INDEX(0, i, KB, TILE_SIZE), m1);
+                acc_C::apply(C + TILE_M * ldc + TILE_N, ldc, Sumi7, &A[TILE_M * KB + i], KB, B + PACKED_INDEX(1, i, KB, TILE_SIZE), m1);
+            }
+        });
+    }
+    return;
+}
+
+} // anonymous namespace
+
+// get the packed tensor size for quantized weights
+size_t ggml_backend_amx_get_alloc_size(const struct ggml_tensor * tensor) {
+    const enum ggml_type TYPE = tensor->type;
+
+    const int K = tensor->ne[0]; // ne0: in_features
+    const int N = tensor->ne[1]; // ne1: out_features
+
+    auto get_tensor_size = [&] {
+        size_t row_size_B{0};
+        GGML_DISPATCH_QTYPES(TYPE, [&] {
+            row_size_B = get_row_size(K);
+        });
+        return N * row_size_B;
+    };
+
+    if (qtype_has_amx_kernels(TYPE)) {
+        return get_tensor_size();
+    } else {
+        // for f16, bf16 we don't do packing
+        return ggml_nbytes(tensor);
+    }
+}
+
+// pack weight to vnni format
+void ggml_backend_amx_convert_weight(struct ggml_tensor * tensor, const void * data, size_t offset, size_t size) {
+
+    size_t alloc_size = ggml_backend_amx_get_alloc_size(tensor);
+    GGML_ASSERT(alloc_size == size);
+
+    const enum ggml_type TYPE = tensor->type;
+
+    const int K = tensor->ne[0]; // ne0: in_features
+    const int N = tensor->ne[1]; // ne1: out_features
+
+#if defined(_OPENMP)
+    // the buffer ctx is not initialized when .set_tensor is called
+    int n_threads = omp_get_num_threads();
+#else
+    int n_threads = 1;
+#endif
+
+    GGML_DISPATCH_QTYPES(TYPE, [&] {
+        convert_B_packed_format((void *)((char *)tensor->data + offset), (const type *)data, N, K, n_threads);
+    });
+}
+
+// NB: mixed dtype gemm with Advanced Matrix Extensions (Intel AMX)
+//
+// src0: weight in shape of {N, K}, quantized
+// src1: input  in shape of {M, K}, float32
+// dst:  output in shape of {M, N}, float32
+//
+// the function performs: dst = src1 @ src0.T
+//
+void ggml_backend_amx_mul_mat(ggml_backend_amx_context * ctx, struct ggml_tensor * dst) {
+    struct ggml_tensor * src0 = dst->src[0];
+    struct ggml_tensor * src1 = dst->src[1];
+
+    const enum ggml_type TYPE = src0->type;
+
+    const int n_threads = ctx->n_threads;
+
+    // f16 only has avx512 kernels for now,
+    // amx kernels will be added once 6th gen xeon is released.
+    const bool is_floating_type = TYPE == GGML_TYPE_F16;
+
+    const int M = dst->ne[1];
+    const int N = dst->ne[0];
+    const int K = src0->ne[0];
+    const int ldc = dst->nb[1] / dst->nb[0];
+
+    if (is_floating_type) {
+        constexpr int BLOCK_M = 4;
+        constexpr int BLOCK_N = 6;
+        const int MB = div_up(M, BLOCK_M);
+        const int NB = div_up(N, BLOCK_N);
+
+        parallel_for(n_threads, MB * NB, [&](int begin, int end) {
+            GGML_DISPATCH_FLOATING_TYPES(TYPE, [&] {
+                for (int i = begin; i < end; ++i) {
+                    int mb = i / NB;
+                    int nb = i % NB;
+
+                    int mb_start = mb * BLOCK_M;
+                    int mb_size = std::min(BLOCK_M, M - mb_start);
+                    int nb_start = nb * BLOCK_N;
+                    int nb_size = std::min(BLOCK_N, N - nb_start);
+
+                    switch (mb_size << 4 | nb_size) {
+                        case 0x12: LAUNCH_TINYGEMM_KERNEL_AVX(1, 2); break;
+                        case 0x14: LAUNCH_TINYGEMM_KERNEL_AVX(1, 4); break;
+                        case 0x16: LAUNCH_TINYGEMM_KERNEL_AVX(1, 6); break;
+                        case 0x22: LAUNCH_TINYGEMM_KERNEL_AVX(2, 2); break;
+                        case 0x24: LAUNCH_TINYGEMM_KERNEL_AVX(2, 4); break;
+                        case 0x26: LAUNCH_TINYGEMM_KERNEL_AVX(2, 6); break;
+                        case 0x32: LAUNCH_TINYGEMM_KERNEL_AVX(3, 2); break;
+                        case 0x34: LAUNCH_TINYGEMM_KERNEL_AVX(3, 4); break;
+                        case 0x36: LAUNCH_TINYGEMM_KERNEL_AVX(3, 6); break;
+                        case 0x42: LAUNCH_TINYGEMM_KERNEL_AVX(4, 2); break;
+                        case 0x44: LAUNCH_TINYGEMM_KERNEL_AVX(4, 4); break;
+                        case 0x46: LAUNCH_TINYGEMM_KERNEL_AVX(4, 6); break;
+                        default: fprintf(stderr, "Unexpected block size!\n");
+                    }
+                }
+            });
+        });
+        return;
+    }
+
+    // pointer to work space, used convert A from float to quantized type
+    void * wdata = nullptr;
+
+    //TODO: performance improvement: merge quant A
+    GGML_DISPATCH_QTYPES(TYPE, [&] {
+        const size_t row_size_A = K / blck_size * sizeof(vec_dot_type);
+        const size_t desired_wsize = M * row_size_A;
+        if (ctx->work_size < desired_wsize) {
+            ctx->work_data.reset(new char[desired_wsize]);
+            ctx->work_size = desired_wsize;
+        }
+        wdata = ctx->work_data.get();
+
+        // Q4_0, Q4_1, Q8_0 handles 1 TILE_K per blck_size
+        // Q4_K, Q5_K, Q6_K, IQ4_XS handles 8 TILE_K per blck_size
+        GGML_ASSERT(TILE_K == blck_size || TILE_K * 8 == blck_size);
+
+        const float * A_data = static_cast(src1->data);
+        for (int m = 0; m < M; ++m) {
+            from_float(A_data + m * K, (char *)wdata + m * row_size_A, K);
+        }
+    });
+
+    if (M == 1) {
+        // MB = 1 and handle 8 tiles in each block
+        constexpr int kTilesN = 4;
+        constexpr int BLOCK_N = TILE_N * kTilesN;
+        const int NB = div_up(N, BLOCK_N);
+
+        parallel_for(n_threads, NB, [&](int begin, int end) {
+            GGML_DISPATCH_QTYPES(TYPE, [&] {
+                const int KB = K / blck_size;
+                const int TILE_SIZE = get_tile_size();
+                const int row_size_A = KB * sizeof(vec_dot_type);
+                for (int i = begin; i < end; ++i) {
+                    int nb = i;
+                    int nb_start = nb * BLOCK_N;
+                    int nb_size = std::min(BLOCK_N, N - nb_start); // 32, 64, 96
+
+                    switch (nb_size) {
+                        //case 160: LAUNCH_TINYGEMM_KERNEL_VNNI(160); break;
+                        case 128: LAUNCH_TINYGEMM_KERNEL_VNNI(128); break;
+                        case 96: LAUNCH_TINYGEMM_KERNEL_VNNI(96); break;
+                        case 64: LAUNCH_TINYGEMM_KERNEL_VNNI(64); break;
+                        case 32: LAUNCH_TINYGEMM_KERNEL_VNNI(32); break;
+                        default: fprintf(stderr, "Unexpected n block size!\n");
+                    }
+                }
+            });
+        });
+        return;
+    }
+
+    // handle 4 tiles at a tile
+    constexpr int BLOCK_M = TILE_M * 2;
+    constexpr int BLOCK_N = TILE_N * 2;
+    const int MB = div_up(M, BLOCK_M);
+    const int NB = div_up(N, BLOCK_N);
+
+    parallel_for(n_threads, MB * NB, [&](int begin, int end) {
+        // init tile config for each thread
+        ggml_tile_config_init();
+
+        GGML_DISPATCH_QTYPES(TYPE, [&] {
+            const int KB = K / blck_size;
+            const int TILE_SIZE = get_tile_size();
+            const int row_size_A = KB * sizeof(vec_dot_type);
+
+            for (int i = begin; i < end; ++i) {
+                int mb = i / NB;
+                int nb = i % NB;
+
+                int mb_start = mb * BLOCK_M;
+                int mb_size = std::min(BLOCK_M, M - mb_start);
+                int nb_start = nb * BLOCK_N;
+                int nb_size = BLOCK_N;
+
+                tinygemm_kernel_amx(
+                    mb_size, nb_size, KB,
+                    (const char *)wdata + mb_start * row_size_A,
+                    (const char *)src0->data + PACKED_INDEX(nb * 2, 0, KB, TILE_SIZE),
+                    (float *) dst->data + mb_start * N + nb_start, ldc);
+            }
+        });
+    });
+}
+
+#else // if defined(__AMX_INT8__)
+
+void ggml_backend_amx_mul_mat(ggml_backend_amx_context * ctx, struct ggml_tensor * dst) {
+    fprintf(stderr, "GGML is not compiled with AMX support!\n");
+
+    GGML_UNUSED(ctx);
+    GGML_UNUSED(dst);
+}
+
+#endif // if defined(__AMX_INT8__)
diff --git a/ggml/src/ggml-amx/mmq.h b/ggml/src/ggml-amx/mmq.h
new file mode 100644
index 000000000..cf0920620
--- /dev/null
+++ b/ggml/src/ggml-amx/mmq.h
@@ -0,0 +1,17 @@
+#pragma once
+#include "common.h"
+#include 
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+size_t ggml_backend_amx_get_alloc_size(const struct ggml_tensor * tensor);
+
+void ggml_backend_amx_convert_weight(struct ggml_tensor * tensor, const void * data, size_t offset, size_t size);
+
+void ggml_backend_amx_mul_mat(ggml_backend_amx_context * ctx, struct ggml_tensor * dst);
+
+#ifdef __cplusplus
+}
+#endif
diff --git a/ggml/src/ggml-backend.cpp b/ggml/src/ggml-backend.cpp
index a3bc79a46..1c17dde30 100644
--- a/ggml/src/ggml-backend.cpp
+++ b/ggml/src/ggml-backend.cpp
@@ -329,7 +329,6 @@ bool ggml_backend_supports_buft(ggml_backend_t backend, ggml_backend_buffer_type
     if (backend->device) {
         return ggml_backend_dev_supports_buft(backend->device, buft);
     }
-
     return backend->iface.supports_buft(backend, buft);
 }
 
@@ -550,6 +549,14 @@ void * ggml_backend_reg_get_proc_address(ggml_backend_reg_t reg, const char * na
 #include "ggml-rpc.h"
 #endif
 
+#ifndef __AMX_INT8__
+#undef GGML_USE_AMX
+#endif
+
+#ifdef GGML_USE_AMX
+#  include "ggml-amx.h"
+#endif
+
 struct ggml_backend_registry {
     std::vector backends;
     std::vector devices;
@@ -570,6 +577,9 @@ struct ggml_backend_registry {
 #ifdef GGML_USE_RPC
         register_backend(ggml_backend_rpc_reg());
 #endif
+#ifdef GGML_USE_AMX
+        register_backend(ggml_backend_amx_reg());
+#endif
 
         // TODO: sycl, kompute, cann
 
diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c
index 779b38d12..7e24313ed 100644
--- a/ggml/src/ggml.c
+++ b/ggml/src/ggml.c
@@ -23252,6 +23252,14 @@ int ggml_cpu_has_avx512_bf16(void) {
 #endif
 }
 
+int ggml_cpu_has_amx_int8(void) {
+#if defined(__AMX_INT8__)
+    return 1;
+#else
+    return 0;
+#endif
+}
+
 int ggml_cpu_has_fma(void) {
 #if defined(__FMA__)
     return 1;
diff --git a/src/llama.cpp b/src/llama.cpp
index dcb015d12..0025e94b8 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -16,6 +16,14 @@
 #   include "ggml-cann.h"
 #endif
 
+#ifndef __AMX_INT8__
+#undef GGML_USE_AMX
+#endif
+
+#ifdef GGML_USE_AMX
+#  include "ggml-amx.h"
+#endif
+
 // TODO: replace with ggml API call
 #define QK_K 256
 
@@ -3533,6 +3541,7 @@ static size_t llama_get_device_memory(const llama_model & model, int device) {
 #else
     return 1;
 #endif
+
     GGML_UNUSED(model);
     GGML_UNUSED(device);
 }
@@ -7031,7 +7040,14 @@ static bool llm_load_tensors(
 
     // assign cpu layers
     for (int i = 0; i < i_gpu_start; ++i) {
+#ifdef GGML_USE_AMX
+        model.buft_layer[i] = {
+            ggml_backend_amx_buffer_type(),
+            llama_default_buffer_type_cpu(model, true)
+        };
+#else
         model.buft_layer[i] = llama_default_buffer_type_cpu(model, true);
+#endif
     }
 
     if (split_mode == LLAMA_SPLIT_MODE_LAYER) {
@@ -21839,6 +21855,7 @@ const char * llama_print_system_info(void) {
     s += "AVX512_VBMI = " + std::to_string(ggml_cpu_has_avx512_vbmi()) + " | ";
     s += "AVX512_VNNI = " + std::to_string(ggml_cpu_has_avx512_vnni()) + " | ";
     s += "AVX512_BF16 = " + std::to_string(ggml_cpu_has_avx512_bf16()) + " | ";
+    s += "AMX_INT8 = "    + std::to_string(ggml_cpu_has_amx_int8())    + " | ";
     s += "FMA = "         + std::to_string(ggml_cpu_has_fma())         + " | ";
     s += "NEON = "        + std::to_string(ggml_cpu_has_neon())        + " | ";
     s += "SVE = "         + std::to_string(ggml_cpu_has_sve())         + " | ";

From 87421a23e8c60e00a7b227d501e8aab2a1aff7ce Mon Sep 17 00:00:00 2001
From: Ouadie EL FAROUKI 
Date: Fri, 18 Oct 2024 06:46:16 +0100
Subject: [PATCH 35/36] [SYCL] Add SYCL Backend registry, device and Event
 Interfaces (#9705)

* implemented missing SYCL event APIs

* sycl : Added device and backend reg interfaces

* Restructured ggml-sycl.cpp
---
 examples/llama-bench/llama-bench.cpp |    2 +-
 ggml/include/ggml-sycl.h             |   11 +-
 ggml/src/ggml-backend.cpp            |   10 +-
 ggml/src/ggml-sycl.cpp               | 2689 ++++++++++++++------------
 src/llama.cpp                        |   61 +-
 5 files changed, 1492 insertions(+), 1281 deletions(-)

diff --git a/examples/llama-bench/llama-bench.cpp b/examples/llama-bench/llama-bench.cpp
index c22bdedcf..60a7aef5b 100644
--- a/examples/llama-bench/llama-bench.cpp
+++ b/examples/llama-bench/llama-bench.cpp
@@ -151,7 +151,7 @@ static std::string get_gpu_info() {
     int count = ggml_backend_sycl_get_device_count();
     for (int i = 0; i < count; i++) {
         char buf[128];
-        ggml_sycl_get_device_description(i, buf, sizeof(buf));
+        ggml_backend_sycl_get_device_description(i, buf, sizeof(buf));
         id += buf;
         if (i < count - 1) {
             id += "/";
diff --git a/ggml/include/ggml-sycl.h b/ggml/include/ggml-sycl.h
index 03b698e61..af521f599 100644
--- a/ggml/include/ggml-sycl.h
+++ b/ggml/include/ggml-sycl.h
@@ -19,6 +19,8 @@ extern "C" {
 // backend API
 GGML_API ggml_backend_t ggml_backend_sycl_init(int device);
 
+GGML_API bool ggml_backend_is_sycl(ggml_backend_t backend);
+
 // devide buffer
 GGML_API ggml_backend_buffer_type_t ggml_backend_sycl_buffer_type(int device);
 
@@ -29,14 +31,19 @@ GGML_API ggml_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const fl
 GGML_API ggml_backend_buffer_type_t ggml_backend_sycl_host_buffer_type(void);
 
 GGML_API void ggml_backend_sycl_print_sycl_devices(void);
-GGML_API void ggml_sycl_get_gpu_list(int *id_list, int max_len);
-GGML_API void ggml_sycl_get_device_description(int device, char *description, size_t description_size);
+GGML_API void ggml_backend_sycl_get_gpu_list(int *id_list, int max_len);
+GGML_API void ggml_backend_sycl_get_device_description(int device,
+                                                       char *description,
+                                                       size_t description_size);
 GGML_API int  ggml_backend_sycl_get_device_count();
 GGML_API void ggml_backend_sycl_get_device_memory(int device, size_t *free, size_t *total);
 
 // SYCL doesn't support registering host memory, keep here for reference
 // GGML_API bool ggml_backend_sycl_register_host_buffer(void * buffer, size_t size);
 // GGML_API void ggml_backend_sycl_unregister_host_buffer(void * buffer);
+
+GGML_API ggml_backend_reg_t ggml_backend_sycl_reg(void);
+
 #ifdef  __cplusplus
 }
 #endif
diff --git a/ggml/src/ggml-backend.cpp b/ggml/src/ggml-backend.cpp
index 1c17dde30..81d09cd8b 100644
--- a/ggml/src/ggml-backend.cpp
+++ b/ggml/src/ggml-backend.cpp
@@ -537,6 +537,10 @@ void * ggml_backend_reg_get_proc_address(ggml_backend_reg_t reg, const char * na
 #include "ggml-metal.h"
 #endif
 
+#ifdef GGML_USE_SYCL
+#include "ggml-sycl.h"
+#endif
+
 #ifdef GGML_USE_VULKAN
 #include "ggml-vulkan.h"
 #endif
@@ -568,6 +572,9 @@ struct ggml_backend_registry {
 #ifdef GGML_USE_METAL
         register_backend(ggml_backend_metal_reg());
 #endif
+#ifdef GGML_USE_SYCL
+        register_backend(ggml_backend_sycl_reg());
+#endif
 #ifdef GGML_USE_VULKAN
         register_backend(ggml_backend_vk_reg());
 #endif
@@ -581,7 +588,7 @@ struct ggml_backend_registry {
         register_backend(ggml_backend_amx_reg());
 #endif
 
-        // TODO: sycl, kompute, cann
+        // TODO: kompute, cann
 
         register_backend(ggml_backend_cpu_reg());
     }
@@ -2254,6 +2261,7 @@ ggml_backend_sched_t ggml_backend_sched_new(
         sched->backends[b] = backends[b];
         sched->bufts[b] = bufts ? bufts[b] : ggml_backend_get_default_buffer_type(backends[b]);
         GGML_ASSERT(ggml_backend_supports_buft(backends[b], sched->bufts[b]));
+
         if (sched->n_copies > 1) {
             for (int c = 0; c < sched->n_copies; c++) {
                 sched->events[b][c] = ggml_backend_event_new(backends[b]->device);
diff --git a/ggml/src/ggml-sycl.cpp b/ggml/src/ggml-sycl.cpp
index 4d3f1c5ce..4d91ee460 100644
--- a/ggml/src/ggml-sycl.cpp
+++ b/ggml/src/ggml-sycl.cpp
@@ -40,17 +40,316 @@
 #include "ggml-sycl/presets.hpp"
 #include "ggml-sycl/gemm.hpp"
 
-bool   ggml_sycl_loaded(void);
-void   ggml_sycl_free_data(struct ggml_tensor * tensor);
-void   ggml_sycl_copy_to_device(struct ggml_tensor * tensor);
-void   ggml_sycl_set_main_device(int main_device);
-void   ggml_sycl_set_mul_mat_q(bool mul_mat_q);
-void   ggml_sycl_get_device_description(int device, char * description, size_t description_size);
-bool   ggml_backend_is_sycl(ggml_backend_t backend);
-int    ggml_backend_sycl_get_device(ggml_backend_t backend);
-static bool ggml_backend_buffer_is_sycl_split(ggml_backend_buffer_t buffer);
-static inline int get_sycl_env(const char *env_name, int default_val);
+static bool g_sycl_loaded = false;
 
+static ggml_sycl_device_info ggml_sycl_init() {
+    ggml_sycl_device_info info = {};
+
+    info.device_count = dpct::dev_mgr::instance().device_count();
+    if (info.device_count == 0) {
+        fprintf(stderr, "%s: failed to initialize " GGML_SYCL_NAME ": %s\n", __func__);
+        return info;
+    }
+
+    GGML_ASSERT(info.device_count <= GGML_SYCL_MAX_DEVICES);
+
+    int64_t total_vram = 0;
+#if defined(GGML_SYCL_FORCE_MMQ)
+    fprintf(stderr, "%s: GGML_SYCL_FORCE_MMQ:   yes\n", __func__);
+#else
+    fprintf(stderr, "%s: GGML_SYCL_FORCE_MMQ:   no\n", __func__);
+#endif
+#if defined(SYCL_USE_XMX)
+    fprintf(stderr, "%s: SYCL_USE_XMX: yes\n", __func__);
+#else
+    fprintf(stderr, "%s: SYCL_USE_XMX: no\n", __func__);
+#endif
+    fprintf(stderr, "%s: found %d " GGML_SYCL_NAME " devices:\n", __func__, info.device_count);
+
+    for (int i = 0; i < info.device_count; ++i) {
+        info.devices[i].vmm = 0;
+        dpct::device_info prop;
+        SYCL_CHECK(CHECK_TRY_ERROR(dpct::get_device_info(
+            prop, dpct::dev_mgr::instance().get_device(i))));
+
+        info.default_tensor_split[i] = total_vram;
+        total_vram += prop.get_global_mem_size();
+
+        info.devices[i].cc =
+            100 * prop.get_major_version() + 10 * prop.get_minor_version();
+
+        info.max_work_group_sizes[i] = prop.get_max_work_group_size();
+    }
+
+    for (int id = 0; id < info.device_count; ++id) {
+        info.default_tensor_split[id] /= total_vram;
+    }
+    return info;
+}
+
+const ggml_sycl_device_info & ggml_sycl_info() {
+    static ggml_sycl_device_info info = ggml_sycl_init();
+    return info;
+}
+
+void print_device_detail(int id, sycl::device &device, std::string device_type) {
+
+    dpct::device_info prop;
+    SYCL_CHECK(CHECK_TRY_ERROR(
+        dpct::get_device_info(prop, device)));
+
+    std::string version;
+    version += std::to_string(prop.get_major_version());
+    version += ".";
+    version += std::to_string(prop.get_minor_version());
+
+    device_type = std::regex_replace(device_type, std::regex("ext_oneapi_"), "");
+    std::string name = std::string(prop.get_name());
+    name = std::regex_replace(name, std::regex("\\(R\\)"), "");
+    name = std::regex_replace(name, std::regex("\\(TM\\)"), "");
+
+    auto global_mem_size = prop.get_global_mem_size()/1000000;
+
+    fprintf(stderr, "|%2d|%19s|%39s|%7s|%7d|%8d|%5d|%6luM|%21s|\n", id, device_type.c_str(),
+            name.c_str(), version.c_str(), prop.get_max_compute_units(),
+            prop.get_max_work_group_size(), prop.get_max_sub_group_size(),
+            global_mem_size, device.get_info().c_str());
+}
+
+void ggml_backend_sycl_print_sycl_devices() {
+    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_print_sycl_devices\n");
+    int device_count = dpct::dev_mgr::instance().device_count();
+    std::map DeviceNums;
+    fprintf(stderr, "found %d SYCL devices:\n", device_count);
+    fprintf(stderr, "|  |                   |                                       |       |Max    |        |Max  |Global |                     |\n");
+    fprintf(stderr, "|  |                   |                                       |       |compute|Max work|sub  |mem    |                     |\n");
+    fprintf(stderr, "|ID|        Device Type|                                   Name|Version|units  |group   |group|size   |       Driver version|\n");
+    fprintf(stderr, "|--|-------------------|---------------------------------------|-------|-------|--------|-----|-------|---------------------|\n");
+    for (int id = 0; id < device_count; ++id) {
+        sycl::device device = dpct::dev_mgr::instance().get_device(id);
+        sycl::backend backend = device.get_backend();
+        std::string backend_type = get_device_backend_and_type(device);
+        int type_id=DeviceNums[backend_type]++;
+        std::stringstream device_type;
+        device_type << "[" <<  backend_type << ":" << std::to_string(type_id) << "]";
+        print_device_detail(id, device, device_type.str());
+    }
+}
+
+static inline int get_sycl_env(const char *env_name, int default_val) {
+    char *user_device_string = getenv(env_name);
+    int user_number = default_val;
+
+    unsigned n;
+    if (user_device_string != NULL &&
+        sscanf(user_device_string, " %u", &n) == 1) {
+        user_number = (int)n;
+    } else {
+        user_number = default_val;
+    }
+    return user_number;
+}
+
+static void ggml_check_sycl() try {
+    static bool initialized = false;
+
+    if (!initialized) {
+        fprintf(stderr, "[SYCL] call ggml_check_sycl\n");
+        g_ggml_sycl_debug = get_sycl_env("GGML_SYCL_DEBUG", 0);
+
+        fprintf(stderr, "%s: GGML_SYCL_DEBUG: %d\n", __func__, g_ggml_sycl_debug);
+
+#if defined(GGML_SYCL_F16)
+        fprintf(stderr, "%s: GGML_SYCL_F16: yes\n", __func__);
+#else
+        fprintf(stderr, "%s: GGML_SYCL_F16: no\n", __func__);
+#endif
+
+/* NOT REMOVE, keep it for next optimize for XMX.
+#if defined(SYCL_USE_XMX)
+        fprintf(stderr, "%s: SYCL_USE_XMX: yes\n", __func__);
+#else
+        fprintf(stderr, "%s: SYCL_USE_XMX: no\n", __func__);
+#endif
+*/
+
+        if (CHECK_TRY_ERROR(g_all_sycl_device_count =
+                            dpct::dev_mgr::instance().device_count()) != 0) {
+            initialized = true;
+            g_sycl_loaded = false;
+            return;
+        }
+        GGML_ASSERT(g_all_sycl_device_count <= GGML_SYCL_MAX_DEVICES);
+        ggml_backend_sycl_print_sycl_devices();
+        initialized = true;
+        g_sycl_loaded = true;
+    }
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+/*
+device_index: device index from 0 to n (continue numbers).
+    It is used for device select/set in SYCL backend internal data structure.
+*/
+inline void check_allow_gpu_index(const int device_index) {
+  if (device_index >= ggml_sycl_info().device_count) {
+    char error_buf[256];
+    snprintf(
+        error_buf,
+        sizeof(error_buf),
+        "%s error: device_index:%d is out of range: [0-%d]",
+        __func__,
+        device_index,
+        ggml_sycl_info().device_count - 1);
+    fprintf(stderr, "%s\n", error_buf);
+    assert(false);
+  }
+}
+
+GGML_API void ggml_backend_sycl_get_gpu_list(int *id_list, int max_len) try {
+    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_get_gpu_list\n");
+    for(int i=0;i=max_len) break;
+        id_list[i] = i;
+    }
+    return;
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+// sycl buffer
+
+struct ggml_backend_sycl_buffer_context {
+    int device;
+    void * dev_ptr = nullptr;
+    queue_ptr stream;
+    std::string name;
+
+     ggml_backend_sycl_buffer_context(int device, void * dev_ptr, queue_ptr stream) :
+        device(device), dev_ptr(dev_ptr), stream(stream) {
+            check_allow_gpu_index(device);
+            name = (GGML_SYCL_NAME + std::to_string(device));
+        }
+
+
+    ~ggml_backend_sycl_buffer_context() {
+        if (dev_ptr != nullptr) {
+            ggml_sycl_set_device(device);
+            SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(dev_ptr, *stream)));
+        }
+    }
+};
+
+static const char * ggml_backend_sycl_buffer_get_name(ggml_backend_buffer_t buffer) {
+    ggml_backend_sycl_buffer_context * ctx = (ggml_backend_sycl_buffer_context *)buffer->context;
+    return ctx->name.c_str();
+}
+
+static bool ggml_backend_buffer_is_sycl(ggml_backend_buffer_t buffer) {
+    return buffer->iface.get_name == ggml_backend_sycl_buffer_get_name;
+}
+
+static void
+ggml_backend_sycl_buffer_free_buffer(ggml_backend_buffer_t buffer) try {
+    ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
+    ggml_sycl_set_device(ctx->device);
+
+    delete ctx;
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+static void * ggml_backend_sycl_buffer_get_base(ggml_backend_buffer_t buffer) {
+    ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
+    return ctx->dev_ptr;
+}
+
+static void
+ggml_backend_sycl_buffer_init_tensor(ggml_backend_buffer_t buffer,
+                                     ggml_tensor *tensor) try {
+    ggml_backend_sycl_buffer_context * ctx = (ggml_backend_sycl_buffer_context *)buffer->context;
+
+    if (tensor->view_src != NULL && tensor->view_offs == 0) {
+        assert(tensor->view_src->buffer->buft == buffer->buft);
+        tensor->backend = tensor->view_src->backend;
+        tensor->extra = tensor->view_src->extra;
+        return;
+    }
+
+
+    if (ggml_is_quantized(tensor->type)) {
+        // initialize padding to 0 to avoid possible NaN values
+        size_t original_size = ggml_nbytes(tensor);
+        size_t padded_size = ggml_backend_buft_get_alloc_size(buffer->buft, tensor);
+
+        if (padded_size > original_size && tensor->view_src == nullptr) {
+            SYCL_CHECK(CHECK_TRY_ERROR(ctx->stream->memset(
+                (char *)tensor->data + original_size, 0,
+                padded_size - original_size).wait()));
+        }
+    }
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+static void ggml_backend_sycl_buffer_set_tensor(ggml_backend_buffer_t buffer,
+                                                ggml_tensor *tensor,
+                                                const void *data, size_t offset,
+                                                size_t size) try {
+
+    ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
+
+    ggml_sycl_set_device(ctx->device);
+    auto stream = &(dpct::dev_mgr::instance().get_device(ctx->device).default_queue());
+    SYCL_CHECK(
+        CHECK_TRY_ERROR(dpct::dev_mgr::instance().get_device(ctx->device).queues_wait_and_throw()));
+    char* host_buf = (char*)malloc(size);
+    memcpy(host_buf, data, size);
+    SYCL_CHECK(
+        CHECK_TRY_ERROR((*stream).memcpy((char *)tensor->data + offset, host_buf, size)
+                             .wait()));
+    free(host_buf);
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+static void ggml_backend_sycl_buffer_get_tensor(ggml_backend_buffer_t buffer,
+                                                const ggml_tensor *tensor,
+                                                void *data, size_t offset,
+                                                size_t size) try {
+
+    ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
+
+    ggml_sycl_set_device(ctx->device);
+    auto stream = dpct::dev_mgr::instance().get_device(ctx->device).default_queue();
+
+    SYCL_CHECK(CHECK_TRY_ERROR(
+        stream.memcpy(data, (const char *)tensor->data + offset, size)
+            .wait()));
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
 
 void dev2dev_memcpy(sycl::queue &q_dst, sycl::queue &q_src, void *ptr_dst,
                     const void *ptr_src, size_t size) {
@@ -60,6 +359,850 @@ void dev2dev_memcpy(sycl::queue &q_dst, sycl::queue &q_src, void *ptr_dst,
     free(host_buf);
 }
 
+static bool
+ggml_backend_sycl_buffer_cpy_tensor(ggml_backend_buffer_t buffer,
+                                    const ggml_tensor *src,
+                                    ggml_tensor *dst) try {
+    if (ggml_backend_buffer_is_sycl(src->buffer)) {
+        ggml_backend_sycl_buffer_context * src_ctx = (ggml_backend_sycl_buffer_context *)src->buffer->context;
+        ggml_backend_sycl_buffer_context * dst_ctx = (ggml_backend_sycl_buffer_context *)dst->buffer->context;
+
+        ggml_sycl_set_device(src_ctx->device);
+        /*
+        DPCT1009:198: SYCL uses exceptions to report errors and does not use the
+        error codes. The original code was commented out and a warning string
+        was inserted. You need to rewrite this code.
+        */
+        SYCL_CHECK(CHECK_TRY_ERROR(
+            dpct::dev_mgr::instance().get_device(src_ctx->device).queues_wait_and_throw()));
+        ggml_sycl_set_device(dst_ctx->device);
+        /*
+        DPCT1009:199: SYCL uses exceptions to report errors and does not use the
+        error codes. The original code was commented out and a warning string
+        was inserted. You need to rewrite this code.
+        */
+        SYCL_CHECK(CHECK_TRY_ERROR(
+            dpct::dev_mgr::instance().get_device(dst_ctx->device).queues_wait_and_throw()));
+        /*
+        DPCT1009:200: SYCL uses exceptions to report errors and does not use the
+        error codes. The original code was commented out and a warning string
+        was inserted. You need to rewrite this code.
+        */
+
+        queue_ptr stream_dst = dst_ctx->stream;
+        queue_ptr stream_src = src_ctx->stream;
+        size_t size = ggml_nbytes(src);
+
+        //todo. it's dirty solutino to walkaroud known issue:device2device cross GPUs.
+        dev2dev_memcpy(*stream_dst, *stream_src, dst->data, src->data, size);
+
+//todo, it's known issue：error in device2device cross GPUs. reused when the issue is fixed. DON"T remove
+#if 0
+        SYCL_CHECK(CHECK_TRY_ERROR((*stream).memcpy(
+            (char *)dst->data, (const char *)src->data, size).wait()));
+
+        /*
+        DPCT1009:201: SYCL uses exceptions to report errors and does not use the
+        error codes. The original code was commented out and a warning string
+        was inserted. You need to rewrite this code.
+        */
+        SYCL_CHECK(CHECK_TRY_ERROR(
+            dpct::dev_mgr::instance().get_device(dst_ctx->device).queues_wait_and_throw()));
+#endif
+        return true;
+    }
+    return false;
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+
+static void ggml_backend_sycl_buffer_clear(ggml_backend_buffer_t buffer,
+                                           uint8_t value) try {
+     ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
+
+    ggml_sycl_set_device(ctx->device);
+    queue_ptr stream = ctx->stream;
+    SYCL_CHECK(
+        CHECK_TRY_ERROR(dpct::get_current_device().queues_wait_and_throw()));
+
+    SYCL_CHECK(CHECK_TRY_ERROR((*stream)
+                                    .memset(ctx->dev_ptr, value, buffer->size)
+                                    .wait()));
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+static const ggml_backend_buffer_i ggml_backend_sycl_buffer_interface = {
+    /* .get_name        = */ ggml_backend_sycl_buffer_get_name,
+    /* .free_buffer     = */ ggml_backend_sycl_buffer_free_buffer,
+    /* .get_base        = */ ggml_backend_sycl_buffer_get_base,
+    /* .init_tensor     = */ ggml_backend_sycl_buffer_init_tensor,
+    /* .memset_tensor   = */ NULL,
+    /* .set_tensor      = */ ggml_backend_sycl_buffer_set_tensor,
+    /* .get_tensor      = */ ggml_backend_sycl_buffer_get_tensor,
+    /* .cpy_tensor      = */ ggml_backend_sycl_buffer_cpy_tensor,
+    /* .clear           = */ ggml_backend_sycl_buffer_clear,
+    /* .reset           = */ NULL,
+};
+
+// sycl buffer type
+struct ggml_backend_sycl_buffer_type_context {
+    int device;
+    std::string name;
+
+    // each buffer type has its own stream
+    queue_ptr stream = nullptr;
+};
+
+static const char * ggml_backend_sycl_buffer_type_get_name(ggml_backend_buffer_type_t buft) {
+    ggml_backend_sycl_buffer_type_context * ctx = (ggml_backend_sycl_buffer_type_context *)buft->context;
+
+    return ctx->name.c_str();
+}
+
+static ggml_backend_buffer_t
+ggml_backend_sycl_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft,
+                                           size_t size) try {
+    ggml_backend_sycl_buffer_type_context * buft_ctx = (ggml_backend_sycl_buffer_type_context *)buft->context;
+    ggml_sycl_set_device(buft_ctx->device);
+    const queue_ptr stream = buft_ctx->stream;
+    size = std::max(size, (size_t)1); // syclMalloc returns null for size 0
+
+    void * dev_ptr;
+    SYCL_CHECK(CHECK_TRY_ERROR(dev_ptr = (void *)sycl::malloc_device(
+                                    size, *stream)));
+    if (!dev_ptr) {
+        fprintf(stderr, "%s: can't malloc %lu Bytes memory on device", __func__, size);
+        return nullptr;
+    }
+    ggml_backend_sycl_buffer_context * ctx = new  ggml_backend_sycl_buffer_context(buft_ctx->device, dev_ptr, buft_ctx->stream);
+    return ggml_backend_buffer_init(buft, ggml_backend_sycl_buffer_interface, ctx, size);
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+static size_t ggml_backend_sycl_buffer_type_get_alignment(ggml_backend_buffer_type_t buft) {
+    return 128;
+    GGML_UNUSED(buft);
+}
+
+static size_t ggml_backend_sycl_buffer_type_get_max_size(ggml_backend_buffer_type_t buft) {
+    return dpct::get_current_device().get_max_mem_alloc_size();
+
+    GGML_UNUSED(buft);
+}
+
+static size_t ggml_backend_sycl_buffer_type_get_alloc_size(ggml_backend_buffer_type_t buft, const ggml_tensor * tensor) {
+    size_t size = ggml_nbytes(tensor);
+    int64_t ne0 = tensor->ne[0];
+
+    if (ggml_is_quantized(tensor->type)) {
+        if (ne0 % MATRIX_ROW_PADDING != 0) {
+            size += ggml_row_size(tensor->type, MATRIX_ROW_PADDING - ne0 % MATRIX_ROW_PADDING);
+        }
+    }
+
+    return size;
+
+    GGML_UNUSED(buft);
+}
+
+static const ggml_backend_buffer_type_i ggml_backend_sycl_buffer_type_interface = {
+    /* .get_name         = */ ggml_backend_sycl_buffer_type_get_name,
+    /* .alloc_buffer     = */ ggml_backend_sycl_buffer_type_alloc_buffer,
+    /* .get_alignment    = */ ggml_backend_sycl_buffer_type_get_alignment,
+    /* .get_max_size     = */ ggml_backend_sycl_buffer_type_get_max_size,
+    /* .get_alloc_size   = */ ggml_backend_sycl_buffer_type_get_alloc_size,
+    /* .is_host          = */ NULL,
+};
+
+ggml_backend_buffer_type_t ggml_backend_sycl_buffer_type(int device) {
+    static std::mutex mutex;
+    std::lock_guard lock(mutex);
+
+    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_buffer_type\n");
+
+    auto dev_count = ggml_backend_sycl_get_device_count();
+
+    if (device>=dev_count or device<0) {
+        printf("ggml_backend_sycl_buffer_type error: device_index:%d is out of range [0, %d], miss to call ggml_backend_sycl_set_single_device()\n",
+            device, dev_count-1);
+        GGML_ASSERT(devicedevice;
+    if (device>=ggml_sycl_info().device_count or device<0) {
+        printf("ggml_backend_sycl_buffer_type error: device_index:%d is out of range [0, %d], miss to call ggml_backend_sycl_set_single_device()\n",
+            device, ggml_sycl_info().device_count-1);
+        GGML_ASSERT(devicestream(i, 0)},
+            };
+        }
+        ggml_backend_sycl_buffer_type_initialized = true;
+    }
+    return &ggml_backend_sycl_buffer_types[device];
+}
+
+// sycl split buffer
+
+static int64_t get_row_rounding(ggml_type type, const std::array & tensor_split) {
+    int64_t min_compute_capability = INT_MAX;
+    int64_t max_compute_capability = INT_MIN;
+    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
+        if (tensor_split[i] < (i + 1 < ggml_sycl_info().device_count ? tensor_split[i + 1] : 1.0f)) {
+            if (min_compute_capability > ggml_sycl_info().devices[i].cc) {
+                min_compute_capability = ggml_sycl_info().devices[i].cc;
+            }
+            if (max_compute_capability < ggml_sycl_info().devices[i].cc) {
+                max_compute_capability = ggml_sycl_info().devices[i].cc;
+            }
+        }
+    }
+
+    switch(type) {
+        case GGML_TYPE_Q4_0:
+        case GGML_TYPE_Q4_1:
+            return max_compute_capability >= VER_GEN9 ? 128 : 64;
+        case GGML_TYPE_Q5_0:
+        case GGML_TYPE_Q5_1:
+        case GGML_TYPE_Q8_0:
+            return 64;
+        case GGML_TYPE_F16:
+        case GGML_TYPE_F32:
+            return 1;
+        case GGML_TYPE_Q2_K:
+        case GGML_TYPE_Q3_K:
+        case GGML_TYPE_Q4_K:
+        case GGML_TYPE_Q5_K:
+        case GGML_TYPE_IQ2_XXS:
+        case GGML_TYPE_IQ2_XS:
+        case GGML_TYPE_IQ2_S:
+        case GGML_TYPE_IQ1_S:
+        case GGML_TYPE_IQ1_M:
+        case GGML_TYPE_IQ3_XXS:
+        case GGML_TYPE_IQ4_XS:
+        case GGML_TYPE_IQ4_NL:
+            return max_compute_capability >= VER_GEN9 ? 128 : 64;
+        case GGML_TYPE_IQ3_S:
+            return max_compute_capability >= VER_GEN9 ? 128 : 64;
+        case GGML_TYPE_Q6_K:
+            return 64;
+        default:
+            GGML_ABORT("fatal error");
+    }
+}
+
+static void get_row_split(int64_t * row_low, int64_t * row_high, const ggml_tensor * tensor, const std::array & tensor_split, int id) {
+    const int64_t nrows = ggml_nrows(tensor);
+    const int64_t rounding = get_row_rounding(tensor->type, tensor_split);
+
+    *row_low = id == 0 ? 0 : nrows*tensor_split[id];
+    *row_low -= *row_low % rounding;
+    if (id == ggml_sycl_info().device_count - 1) {
+        *row_high = nrows;
+    } else {
+        *row_high = nrows*tensor_split[id + 1];
+        *row_high -= *row_high % rounding;
+    }
+}
+
+static size_t ggml_nbytes_split(const struct ggml_tensor * tensor, int nrows_split) {
+    static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
+
+    return nrows_split*ggml_row_size(tensor->type, tensor->ne[0]);
+}
+
+struct ggml_backend_sycl_split_buffer_type_context {
+    std::array tensor_split;
+};
+
+struct ggml_backend_sycl_split_buffer_context {
+    ~ggml_backend_sycl_split_buffer_context() try {
+        for (ggml_tensor_extra_gpu * extra : tensor_extras) {
+            for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
+                for (int64_t is = 0; is < GGML_SYCL_MAX_STREAMS; ++is) {
+                    if (extra->events[i][is] != nullptr) {
+                        /*
+                        DPCT1009:206: SYCL uses exceptions to report errors and
+                        does not use the error codes. The original code was
+                        commented out and a warning string was inserted. You
+                        need to rewrite this code.
+                        */
+                        SYCL_CHECK(CHECK_TRY_ERROR(
+                            dpct::destroy_event(extra->events[i][is])));
+                    }
+                }
+                if (extra->data_device[i] != nullptr) {
+                    /*
+                    DPCT1009:207: SYCL uses exceptions to report errors and does
+                    not use the error codes. The original code was commented out
+                    and a warning string was inserted. You need to rewrite this
+                    code.
+                    */
+                    ggml_sycl_set_device(i);
+                    SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(
+                        extra->data_device[i], *(streams[i]))));
+                }
+            }
+            delete extra;
+        }
+    }
+    catch (sycl::exception const &exc) {
+      std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+                << ", line:" << __LINE__ << std::endl;
+      std::exit(1);
+    }
+
+    std::vector tensor_extras;
+    std::vector streams;
+};
+
+static const char * ggml_backend_sycl_split_buffer_get_name(ggml_backend_buffer_t buffer) {
+    return GGML_SYCL_NAME "_Split";
+
+    GGML_UNUSED(buffer);
+}
+
+static bool ggml_backend_buffer_is_sycl_split(ggml_backend_buffer_t buffer) {
+   return buffer->iface.get_name == ggml_backend_sycl_split_buffer_get_name;
+}
+
+static void ggml_backend_sycl_split_buffer_free_buffer(ggml_backend_buffer_t buffer) {
+    ggml_backend_sycl_split_buffer_context * ctx = (ggml_backend_sycl_split_buffer_context *)buffer->context;
+    delete ctx;
+}
+
+static void * ggml_backend_sycl_split_buffer_get_base(ggml_backend_buffer_t buffer) {
+    // the pointers are stored in the tensor extras, this is just a dummy address and never dereferenced
+    return (void *)0x1000;
+
+    GGML_UNUSED(buffer);
+}
+
+static void
+ggml_backend_sycl_split_buffer_init_tensor(ggml_backend_buffer_t buffer,
+                                           ggml_tensor *tensor) try {
+    GGML_ASSERT(tensor->view_src == nullptr); // views of split tensors are not supported
+
+    ggml_backend_sycl_split_buffer_context * ctx = (ggml_backend_sycl_split_buffer_context *)buffer->context;
+    ggml_backend_sycl_split_buffer_type_context * buft_ctx = (ggml_backend_sycl_split_buffer_type_context *)buffer->buft->context;
+
+    const int64_t ne0 = tensor->ne[0];
+
+    ggml_tensor_extra_gpu * extra = new ggml_tensor_extra_gpu{};
+
+    ctx->tensor_extras.push_back(extra);
+        ctx->streams.push_back(&(dpct::get_current_device().default_queue()));
+
+    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
+        int64_t row_low, row_high;
+        get_row_split(&row_low, &row_high, tensor, buft_ctx->tensor_split, i);
+
+        int64_t nrows_split = row_high - row_low;
+        if (nrows_split == 0) {
+            continue;
+        }
+
+        size_t size = ggml_nbytes_split(tensor, nrows_split);
+        const size_t original_size = size;
+
+        // pad last row to a multiple of 512 elements to avoid out-of-bounds memory accesses
+        if (ne0 % MATRIX_ROW_PADDING != 0) {
+            size += ggml_row_size(tensor->type, MATRIX_ROW_PADDING - ne0 % MATRIX_ROW_PADDING);
+        }
+
+        // FIXME: do not crash if cudaMalloc fails
+        // currently, init_tensor cannot fail, it needs to be fixed in ggml-backend first
+        ggml_sycl_set_device(i);
+        const queue_ptr stream = ctx->streams[i];
+        char * buf;
+        /*
+        DPCT1009:208: SYCL uses exceptions to report errors and does not use the
+        error codes. The original code was commented out and a warning string
+        was inserted. You need to rewrite this code.
+        */
+        SYCL_CHECK(CHECK_TRY_ERROR(buf = (char *)sycl::malloc_device(
+                                        size, *stream)));
+        if (!buf) {
+            char err_buf[1024];
+            snprintf(err_buf, 1023, "%s: can't malloc %lu Bytes memory on device", __func__, size);
+            throw std::runtime_error(err_buf);
+        }
+        // set padding to 0 to avoid possible NaN values
+        if (size > original_size) {
+            /*
+            DPCT1009:209: SYCL uses exceptions to report errors and does not use
+            the error codes. The original code was commented out and a warning
+            string was inserted. You need to rewrite this code.
+            */
+            SYCL_CHECK(CHECK_TRY_ERROR(
+                (*stream)
+                    .memset(buf + original_size, 0, size - original_size)
+                    .wait()));
+        }
+
+        extra->data_device[i] = buf;
+
+        for (int64_t is = 0; is < GGML_SYCL_MAX_STREAMS; ++is) {
+            /*
+            DPCT1009:210: SYCL uses exceptions to report errors and does not use
+            the error codes. The original code was commented out and a warning
+            string was inserted. You need to rewrite this code.
+            */
+            SYCL_CHECK(
+                CHECK_TRY_ERROR(extra->events[i][is] = new sycl::event()));
+        }
+    }
+    tensor->backend = GGML_BACKEND_TYPE_GPU_SPLIT;
+    tensor->extra = extra;
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+static void
+ggml_backend_sycl_split_buffer_set_tensor(ggml_backend_buffer_t buffer,
+                                          ggml_tensor *tensor, const void *data,
+                                          size_t offset, size_t size) try {
+    // split tensors must always be set in their entirety at once
+    GGML_ASSERT(offset == 0);
+    GGML_ASSERT(size == ggml_nbytes(tensor));
+
+    ggml_backend_sycl_split_buffer_context * ctx = (ggml_backend_sycl_split_buffer_context *)buffer->context;
+    ggml_backend_sycl_split_buffer_type_context * buft_ctx = (ggml_backend_sycl_split_buffer_type_context *)buffer->buft->context;
+
+    const int64_t ne0 = tensor->ne[0];
+    const size_t nb1 = tensor->nb[1];
+    ggml_tensor_extra_gpu * extra = (ggml_tensor_extra_gpu *)tensor->extra;
+
+    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
+        int64_t row_low, row_high;
+        get_row_split(&row_low, &row_high, tensor, buft_ctx->tensor_split, i);
+
+        int64_t nrows_split = row_high - row_low;
+        if (nrows_split == 0) {
+            continue;
+        }
+
+        const size_t offset_split = row_low*nb1;
+        size_t size = ggml_nbytes_split(tensor, nrows_split);
+        const size_t original_size = size;
+
+        // pad last row to a multiple of 512 elements to avoid out-of-bounds memory accesses
+        if (ne0 % MATRIX_ROW_PADDING != 0) {
+            size += ggml_row_size(tensor->type, MATRIX_ROW_PADDING - ne0 % MATRIX_ROW_PADDING);
+        }
+
+        const char * buf_host = (const char *)data + offset_split;
+        /*
+        DPCT1009:211: SYCL uses exceptions to report errors and does not use the
+        error codes. The original code was commented out and a warning string
+        was inserted. You need to rewrite this code.
+        */
+        ggml_sycl_set_device(i);
+        const queue_ptr stream = ctx->streams[i];
+        SYCL_CHECK(CHECK_TRY_ERROR(
+            (*stream)
+                .memcpy(extra->data_device[i], buf_host, original_size)
+                .wait()));
+    }
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+static void
+ggml_backend_sycl_split_buffer_get_tensor(ggml_backend_buffer_t buffer,
+                                          const ggml_tensor *tensor, void *data,
+                                          size_t offset, size_t size) try {
+    // split tensors must always be set in their entirety at once
+    GGML_ASSERT(offset == 0);
+    GGML_ASSERT(size == ggml_nbytes(tensor));
+
+    ggml_backend_sycl_split_buffer_context * ctx = (ggml_backend_sycl_split_buffer_context *)buffer->context;
+    ggml_backend_sycl_split_buffer_type_context * buft_ctx = (ggml_backend_sycl_split_buffer_type_context *)buffer->buft->context;
+
+    const int64_t ne0 = tensor->ne[0];
+    const size_t nb1 = tensor->nb[1];
+    ggml_tensor_extra_gpu * extra = (ggml_tensor_extra_gpu *)tensor->extra;
+
+    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
+        int64_t row_low, row_high;
+        get_row_split(&row_low, &row_high, tensor, buft_ctx->tensor_split, i);
+
+        int64_t nrows_split = row_high - row_low;
+        if (nrows_split == 0) {
+            continue;
+        }
+
+        const size_t offset_split = row_low*nb1;
+        size_t size = ggml_nbytes_split(tensor, nrows_split);
+        const size_t original_size = size;
+
+        // pad last row to a multiple of 512 elements to avoid out-of-bounds memory accesses
+        if (ne0 % MATRIX_ROW_PADDING != 0) {
+            size += ggml_row_size(tensor->type, MATRIX_ROW_PADDING - ne0 % MATRIX_ROW_PADDING);
+        }
+
+        char * buf_host = (char *)data + offset_split;
+        /*
+        DPCT1009:212: SYCL uses exceptions to report errors and does not use the
+        error codes. The original code was commented out and a warning string
+        was inserted. You need to rewrite this code.
+        */
+        ggml_sycl_set_device(i);
+        const queue_ptr stream = ctx->streams[i];
+        SYCL_CHECK(CHECK_TRY_ERROR(
+            (*stream)
+                .memcpy(buf_host, extra->data_device[i], original_size)
+                .wait()));
+    }
+}
+catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+static void ggml_backend_sycl_split_buffer_clear(ggml_backend_buffer_t buffer, uint8_t value) {
+    GGML_UNUSED(buffer);
+    GGML_UNUSED(value);
+}
+
+static struct ggml_backend_buffer_i ggml_backend_sycl_split_buffer_interface = {
+    /* .get_name        = */ ggml_backend_sycl_split_buffer_get_name,
+    /* .free_buffer     = */ ggml_backend_sycl_split_buffer_free_buffer,
+    /* .get_base        = */ ggml_backend_sycl_split_buffer_get_base,
+    /* .init_tensor     = */ ggml_backend_sycl_split_buffer_init_tensor,
+    /* .memset_tensor   = */ NULL,
+    /* .set_tensor      = */ ggml_backend_sycl_split_buffer_set_tensor,
+    /* .get_tensor      = */ ggml_backend_sycl_split_buffer_get_tensor,
+    /* .cpy_tensor      = */ NULL,
+    /* .clear           = */ ggml_backend_sycl_split_buffer_clear,
+    /* .reset           = */ NULL,
+};
+
+// sycl split buffer type
+
+static const char * ggml_backend_sycl_split_buffer_type_get_name(ggml_backend_buffer_type_t buft) {
+    return GGML_SYCL_NAME "_Split";
+
+    GGML_UNUSED(buft);
+}
+
+static ggml_backend_buffer_t ggml_backend_sycl_split_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
+    // since we don't know the exact split after rounding, we cannot allocate the device buffers at this point
+    // instead, we allocate them for each tensor separately in init_tensor
+    // however, the size still represents the maximum cumulative size of all the device buffers after the tensors are allocated,
+    // as returned by get_alloc_size. this limit is enforced during tensor allocation by ggml-alloc, so it must be correct.
+    ggml_backend_sycl_split_buffer_context * ctx = new ggml_backend_sycl_split_buffer_context();
+
+    return ggml_backend_buffer_init(buft, ggml_backend_sycl_split_buffer_interface, ctx, size);
+}
+
+static size_t ggml_backend_sycl_split_buffer_type_get_alignment(ggml_backend_buffer_type_t buft) {
+    return 128;
+    GGML_UNUSED(buft);
+}
+
+static size_t ggml_backend_sycl_split_buffer_type_get_alloc_size(ggml_backend_buffer_type_t buft, const ggml_tensor * tensor) {
+    ggml_backend_sycl_split_buffer_type_context * ctx = (ggml_backend_sycl_split_buffer_type_context *)buft->context;
+
+    size_t total_size = 0;
+
+    const int64_t ne0 = tensor->ne[0];
+
+    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
+        int64_t row_low, row_high;
+        get_row_split(&row_low, &row_high, tensor, ctx->tensor_split, i);
+
+        int64_t nrows_split = row_high - row_low;
+        if (nrows_split == 0) {
+            continue;
+        }
+
+        total_size += ggml_nbytes_split(tensor, nrows_split);
+
+        // pad last row to a multiple of 512 elements to avoid out-of-bounds memory accesses
+        if (ne0 % MATRIX_ROW_PADDING != 0) {
+            total_size += ggml_row_size(tensor->type, MATRIX_ROW_PADDING - ne0 % MATRIX_ROW_PADDING);
+        }
+    }
+
+    return total_size;
+}
+
+static bool ggml_backend_sycl_split_buffer_type_is_host(ggml_backend_buffer_type_t buft) {
+    return false;
+
+    GGML_UNUSED(buft);
+}
+
+static ggml_backend_buffer_type_i ggml_backend_sycl_split_buffer_type_interface = {
+    /* .get_name         = */ ggml_backend_sycl_split_buffer_type_get_name,
+    /* .alloc_buffer     = */ ggml_backend_sycl_split_buffer_type_alloc_buffer,
+    /* .get_alignment    = */ ggml_backend_sycl_split_buffer_type_get_alignment,
+    /* .get_max_size     = */ NULL, // defaults to SIZE_MAX
+    /* .get_alloc_size   = */ ggml_backend_sycl_split_buffer_type_get_alloc_size,
+    /* .is_host          = */ ggml_backend_sycl_split_buffer_type_is_host,
+};
+
+ggml_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const float * tensor_split) {
+    static std::mutex mutex;
+    std::lock_guard lock(mutex);
+
+    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_split_buffer_type\n");
+    ggml_check_sycl();
+    // FIXME: this is not thread safe
+    static std::map, struct ggml_backend_buffer_type> buft_map;
+
+    std::array tensor_split_arr = {};
+
+    bool all_zero = tensor_split == nullptr || std::all_of(tensor_split, tensor_split + GGML_SYCL_MAX_DEVICES, [](float x) { return x == 0.0f; });
+    if (all_zero) {
+        tensor_split_arr = ggml_sycl_info().default_tensor_split;
+    } else {
+        float split_sum = 0.0f;
+        for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
+            tensor_split_arr[i] = split_sum;
+            split_sum += tensor_split[i];
+        }
+        for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
+            tensor_split_arr[i] /= split_sum;
+        }
+    }
+
+    auto it = buft_map.find(tensor_split_arr);
+    if (it != buft_map.end()) {
+        return &it->second;
+    }
+
+    struct ggml_backend_buffer_type buft {
+        /* .iface   = */ ggml_backend_sycl_split_buffer_type_interface,
+        /* .device  = */ ggml_backend_reg_dev_get(ggml_backend_sycl_reg(), 0),
+        /* .context = */ new ggml_backend_sycl_split_buffer_type_context{tensor_split_arr},
+    };
+
+    auto result = buft_map.emplace(tensor_split_arr, buft);
+    return &result.first->second;
+}
+
+// host buffer type
+
+static const char * ggml_backend_sycl_host_buffer_type_name(ggml_backend_buffer_type_t buft) {
+    return GGML_SYCL_NAME "_Host";
+
+    GGML_UNUSED(buft);
+}
+
+static const char * ggml_backend_sycl_host_buffer_name(ggml_backend_buffer_t buffer) {
+    return GGML_SYCL_NAME "_Host";
+
+    GGML_UNUSED(buffer);
+}
+
+static void ggml_backend_sycl_host_buffer_free_buffer(ggml_backend_buffer_t buffer) {
+    ggml_sycl_host_free(buffer->context);
+}
+
+static ggml_backend_buffer_t ggml_backend_sycl_host_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
+    void * ptr = ggml_sycl_host_malloc(size);
+
+    if (ptr == nullptr) {
+        // fallback to cpu buffer
+        return ggml_backend_buft_alloc_buffer(ggml_backend_cpu_buffer_type(), size);
+    }
+
+    // FIXME: this is a hack to avoid having to implement a new buffer type
+    ggml_backend_buffer_t buffer = ggml_backend_cpu_buffer_from_ptr(ptr, size);
+    buffer->buft = buft;
+    buffer->iface.get_name = ggml_backend_sycl_host_buffer_name;
+    buffer->iface.free_buffer = ggml_backend_sycl_host_buffer_free_buffer;
+
+    return buffer;
+}
+
+ggml_backend_buffer_type_t ggml_backend_sycl_host_buffer_type() {
+    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_host_buffer_type\n");
+    static struct ggml_backend_buffer_type ggml_backend_sycl_buffer_type_host = {
+        /* .iface    = */ {
+            /* .get_name         = */ ggml_backend_sycl_host_buffer_type_name,
+            /* .alloc_buffer     = */ ggml_backend_sycl_host_buffer_type_alloc_buffer,
+            /* .get_alignment    = */ ggml_backend_cpu_buffer_type()->iface.get_alignment,
+            /* .get_max_size     = */ NULL, // TODO: return device.maxBufferLength
+            /* .get_alloc_size   = */ ggml_backend_cpu_buffer_type()->iface.get_alloc_size,
+            /* .is_host          = */ ggml_backend_cpu_buffer_type()->iface.is_host,
+        },
+        /* .device   = */ ggml_backend_reg_dev_get(ggml_backend_sycl_reg(), 0),
+        /* .context  = */ nullptr,
+    };
+
+    return &ggml_backend_sycl_buffer_type_host;
+}
+
+// buffer pool for sycl (legacy)
+struct ggml_sycl_pool_leg : public ggml_sycl_pool {
+    static const int MAX_SYCL_BUFFERS = 256;
+
+    int device;
+    queue_ptr qptr;
+    struct ggml_sycl_buffer {
+        void * ptr = nullptr;
+        size_t size = 0;
+    };
+
+    ggml_sycl_buffer buffer_pool[MAX_SYCL_BUFFERS] = {};
+    size_t pool_size = 0;
+
+    explicit ggml_sycl_pool_leg(queue_ptr qptr_, int device_) :
+        qptr(qptr_),
+        device(device_) {
+    }
+
+    ~ggml_sycl_pool_leg() {
+        for (int i = 0; i < MAX_SYCL_BUFFERS; ++i) {
+            ggml_sycl_buffer & b = buffer_pool[i];
+            if (b.ptr != nullptr) {
+                SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(b.ptr, *qptr)));
+                pool_size -= b.size;
+            }
+        }
+        GGML_ASSERT(pool_size == 0);
+    }
+
+    void * alloc(size_t size, size_t * actual_size) override {
+#ifdef DEBUG_sycl_MALLOC
+        int nnz = 0;
+        size_t max_size = 0;
+#endif
+        size_t best_diff = 1ull << 36;
+        int ibest = -1;
+        for (int i = 0; i < MAX_SYCL_BUFFERS; ++i) {
+            ggml_sycl_buffer& b = buffer_pool[i];
+            if (b.ptr != nullptr) {
+#ifdef DEBUG_sycl_MALLOC
+                ++nnz;
+                if (b.size > max_size) max_size = b.size;
+#endif
+                if (b.size >= size) {
+                    size_t diff = b.size - size;
+                    if (diff < best_diff) {
+                        best_diff = diff;
+                        ibest = i;
+                        if (!best_diff) {
+                            void * ptr = b.ptr;
+                            *actual_size = b.size;
+                            b.ptr = nullptr;
+                            b.size = 0;
+                            return ptr;
+                        }
+                    }
+                }
+            }
+        }
+        if (ibest >= 0) {
+            ggml_sycl_buffer& b = buffer_pool[ibest];
+            void * ptr = b.ptr;
+            *actual_size = b.size;
+            b.ptr = nullptr;
+            b.size = 0;
+            return ptr;
+        }
+        void * ptr;
+        size_t look_ahead_size = (size_t) (1.05 * size);
+
+        SYCL_CHECK(
+            CHECK_TRY_ERROR(ptr = (void *)sycl::malloc_device(
+                                look_ahead_size, *qptr)));
+        if (!ptr) {
+            fprintf(stderr, "%s: can't malloc %lu Bytes memory on device", __func__, look_ahead_size);
+            return nullptr;
+        }
+
+        *actual_size = look_ahead_size;
+        pool_size += look_ahead_size;
+
+    #ifdef DEBUG_SYCL_MALLOC
+        fprintf(stderr, "%s[%d]: %d buffers, max_size = %u MB, pool_size = %u MB, requested %u MB\n", __func__, id, nnz,
+                (uint32_t)(max_size/1024/1024), (uint32_t)(g_sycl_pool_size[id]/1024/1024), (uint32_t)(size/1024/1024));
+    #endif
+        // GGML_SYCL_DEBUG("ggml_sycl_pool_malloc_leg look_ahead_size=%lu, return %p\n", look_ahead_size, ptr);
+        return ptr;
+    }
+
+    void free(void * ptr, size_t size) override {
+        for (int i = 0; i < MAX_SYCL_BUFFERS; ++i) {
+            ggml_sycl_buffer& b = buffer_pool[i];
+            if (b.ptr == nullptr) {
+                b.ptr = ptr;
+                b.size = size;
+                return;
+            }
+        }
+        fprintf(stderr, "WARNING: sycl buffer pool full, increase MAX_sycl_BUFFERS\n");
+        SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(ptr, *qptr)));
+        pool_size -= size;
+    }
+};
+
+std::unique_ptr ggml_backend_sycl_context::new_pool_for_device(queue_ptr qptr, int device) {
+    // TBD: NO VMM support
+    // if (ggml_sycl_info().devices[device].vmm) {
+    //     return std::unique_ptr(new ggml_sycl_pool_vmm(device));
+    // }
+   return std::unique_ptr(new ggml_sycl_pool_leg(qptr, device));
+}
+
+// TBD pool with virtual memory management
+// struct ggml_sycl_pool_vmm : public ggml_sycl_pool
+
+/// kernels
+
 typedef void (*cpy_kernel_t)(const char * cx, char * cdst);
 typedef void (*ggml_sycl_func_t)(ggml_backend_sycl_context & ctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst);
 typedef void (*ggml_sycl_op_mul_mat_t)(
@@ -1706,296 +2849,6 @@ static void diag_mask_inf_f32_sycl(const float *x, float *dst,
                          });
 }
 
-static bool g_sycl_loaded = false;
-
-bool ggml_sycl_loaded(void) {
-    return g_sycl_loaded;
-}
-
-void print_device_detail(int id, sycl::device &device, std::string device_type) {
-
-    dpct::device_info prop;
-    SYCL_CHECK(CHECK_TRY_ERROR(
-        dpct::get_device_info(prop, device)));
-
-    std::string version;
-    version += std::to_string(prop.get_major_version());
-    version += ".";
-    version += std::to_string(prop.get_minor_version());
-
-    device_type = std::regex_replace(device_type, std::regex("ext_oneapi_"), "");
-    std::string name = std::string(prop.get_name());
-    name = std::regex_replace(name, std::regex("\\(R\\)"), "");
-    name = std::regex_replace(name, std::regex("\\(TM\\)"), "");
-
-    auto global_mem_size = prop.get_global_mem_size()/1000000;
-
-    fprintf(stderr, "|%2d|%19s|%39s|%7s|%7d|%8d|%5d|%6luM|%21s|\n", id, device_type.c_str(),
-            name.c_str(), version.c_str(), prop.get_max_compute_units(),
-            prop.get_max_work_group_size(), prop.get_max_sub_group_size(),
-            global_mem_size, device.get_info().c_str());
-}
-
-void ggml_backend_sycl_print_sycl_devices() {
-    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_print_sycl_devices\n");
-    int device_count = dpct::dev_mgr::instance().device_count();
-    std::map DeviceNums;
-    fprintf(stderr, "found %d SYCL devices:\n", device_count);
-    fprintf(stderr, "|  |                   |                                       |       |Max    |        |Max  |Global |                     |\n");
-    fprintf(stderr, "|  |                   |                                       |       |compute|Max work|sub  |mem    |                     |\n");
-    fprintf(stderr, "|ID|        Device Type|                                   Name|Version|units  |group   |group|size   |       Driver version|\n");
-    fprintf(stderr, "|--|-------------------|---------------------------------------|-------|-------|--------|-----|-------|---------------------|\n");
-    for (int id = 0; id < device_count; ++id) {
-        sycl::device device = dpct::dev_mgr::instance().get_device(id);
-        sycl::backend backend = device.get_backend();
-        std::string backend_type = get_device_backend_and_type(device);
-        int type_id=DeviceNums[backend_type]++;
-        std::stringstream device_type;
-        device_type << "[" <<  backend_type << ":" << std::to_string(type_id) << "]";
-        print_device_detail(id, device, device_type.str());
-    }
-}
-
-static inline int get_sycl_env(const char *env_name, int default_val) {
-    char *user_device_string = getenv(env_name);
-    int user_number = default_val;
-
-    unsigned n;
-    if (user_device_string != NULL &&
-        sscanf(user_device_string, " %u", &n) == 1) {
-        user_number = (int)n;
-    } else {
-        user_number = default_val;
-    }
-    return user_number;
-}
-
-static void ggml_check_sycl() try {
-    static bool initialized = false;
-
-    if (!initialized) {
-        fprintf(stderr, "[SYCL] call ggml_check_sycl\n");
-        g_ggml_sycl_debug = get_sycl_env("GGML_SYCL_DEBUG", 0);
-
-        fprintf(stderr, "%s: GGML_SYCL_DEBUG: %d\n", __func__, g_ggml_sycl_debug);
-
-#if defined(GGML_SYCL_F16)
-        fprintf(stderr, "%s: GGML_SYCL_F16: yes\n", __func__);
-#else
-        fprintf(stderr, "%s: GGML_SYCL_F16: no\n", __func__);
-#endif
-
-/* NOT REMOVE, keep it for next optimize for XMX.
-#if defined(SYCL_USE_XMX)
-        fprintf(stderr, "%s: SYCL_USE_XMX: yes\n", __func__);
-#else
-        fprintf(stderr, "%s: SYCL_USE_XMX: no\n", __func__);
-#endif
-*/
-
-        if (CHECK_TRY_ERROR(g_all_sycl_device_count =
-                            dpct::dev_mgr::instance().device_count()) != 0) {
-            initialized = true;
-            g_sycl_loaded = false;
-            return;
-        }
-        GGML_ASSERT(g_all_sycl_device_count <= GGML_SYCL_MAX_DEVICES);
-        ggml_backend_sycl_print_sycl_devices();
-        initialized = true;
-        g_sycl_loaded = true;
-    }
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-static ggml_sycl_device_info ggml_sycl_init() {
-    ggml_sycl_device_info info = {};
-
-    info.device_count = dpct::dev_mgr::instance().device_count();
-    if (info.device_count == 0) {
-        fprintf(stderr, "%s: failed to initialize " GGML_SYCL_NAME ": %s\n", __func__);
-        return info;
-    }
-
-    GGML_ASSERT(info.device_count <= GGML_SYCL_MAX_DEVICES);
-
-    int64_t total_vram = 0;
-#if defined(GGML_SYCL_FORCE_MMQ)
-    fprintf(stderr, "%s: GGML_SYCL_FORCE_MMQ:   yes\n", __func__);
-#else
-    fprintf(stderr, "%s: GGML_SYCL_FORCE_MMQ:   no\n", __func__);
-#endif
-#if defined(SYCL_USE_XMX)
-    fprintf(stderr, "%s: SYCL_USE_XMX: yes\n", __func__);
-#else
-    fprintf(stderr, "%s: SYCL_USE_XMX: no\n", __func__);
-#endif
-    fprintf(stderr, "%s: found %d " GGML_SYCL_NAME " devices:\n", __func__, info.device_count);
-
-    for (int i = 0; i < info.device_count; ++i) {
-        info.devices[i].vmm = 0;
-        dpct::device_info prop;
-        SYCL_CHECK(CHECK_TRY_ERROR(dpct::get_device_info(
-            prop, dpct::dev_mgr::instance().get_device(i))));
-
-        info.default_tensor_split[i] = total_vram;
-        total_vram += prop.get_global_mem_size();
-
-        info.devices[i].cc =
-            100 * prop.get_major_version() + 10 * prop.get_minor_version();
-
-        info.max_work_group_sizes[i] = prop.get_max_work_group_size();
-    }
-
-    for (int id = 0; id < info.device_count; ++id) {
-        info.default_tensor_split[id] /= total_vram;
-    }
-    return info;
-}
-
-const ggml_sycl_device_info & ggml_sycl_info() {
-    static ggml_sycl_device_info info = ggml_sycl_init();
-    return info;
-}
-
-/*
-device_index: device index from 0 to n (continue numbers).
-    It is used for device select/set in SYCL backend internal data structure.
-*/
-inline void check_allow_gpu_index(const int device_index) {
-  if (device_index >= ggml_sycl_info().device_count) {
-    char error_buf[256];
-    snprintf(
-        error_buf,
-        sizeof(error_buf),
-        "%s error: device_index:%d is out of range: [0-%d]",
-        __func__,
-        device_index,
-        ggml_sycl_info().device_count - 1);
-    fprintf(stderr, "%s\n", error_buf);
-    assert(false);
-  }
-}
-
-// buffer pool for sycl (legacy)
-struct ggml_sycl_pool_leg : public ggml_sycl_pool {
-    static const int MAX_SYCL_BUFFERS = 256;
-
-    int device;
-    queue_ptr qptr;
-    struct ggml_sycl_buffer {
-        void * ptr = nullptr;
-        size_t size = 0;
-    };
-
-    ggml_sycl_buffer buffer_pool[MAX_SYCL_BUFFERS] = {};
-    size_t pool_size = 0;
-
-    explicit ggml_sycl_pool_leg(queue_ptr qptr_, int device_) :
-        qptr(qptr_),
-        device(device_) {
-    }
-
-    ~ggml_sycl_pool_leg() {
-        for (int i = 0; i < MAX_SYCL_BUFFERS; ++i) {
-            ggml_sycl_buffer & b = buffer_pool[i];
-            if (b.ptr != nullptr) {
-                SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(b.ptr, *qptr)));
-                pool_size -= b.size;
-            }
-        }
-        GGML_ASSERT(pool_size == 0);
-    }
-
-    void * alloc(size_t size, size_t * actual_size) override {
-#ifdef DEBUG_sycl_MALLOC
-        int nnz = 0;
-        size_t max_size = 0;
-#endif
-        size_t best_diff = 1ull << 36;
-        int ibest = -1;
-        for (int i = 0; i < MAX_SYCL_BUFFERS; ++i) {
-            ggml_sycl_buffer& b = buffer_pool[i];
-            if (b.ptr != nullptr) {
-#ifdef DEBUG_sycl_MALLOC
-                ++nnz;
-                if (b.size > max_size) max_size = b.size;
-#endif
-                if (b.size >= size) {
-                    size_t diff = b.size - size;
-                    if (diff < best_diff) {
-                        best_diff = diff;
-                        ibest = i;
-                        if (!best_diff) {
-                            void * ptr = b.ptr;
-                            *actual_size = b.size;
-                            b.ptr = nullptr;
-                            b.size = 0;
-                            return ptr;
-                        }
-                    }
-                }
-            }
-        }
-        if (ibest >= 0) {
-            ggml_sycl_buffer& b = buffer_pool[ibest];
-            void * ptr = b.ptr;
-            *actual_size = b.size;
-            b.ptr = nullptr;
-            b.size = 0;
-            return ptr;
-        }
-        void * ptr;
-        size_t look_ahead_size = (size_t) (1.05 * size);
-
-        SYCL_CHECK(
-            CHECK_TRY_ERROR(ptr = (void *)sycl::malloc_device(
-                                look_ahead_size, *qptr)));
-        if (!ptr) {
-            fprintf(stderr, "%s: can't malloc %lu Bytes memory on device", __func__, look_ahead_size);
-            return nullptr;
-        }
-
-        *actual_size = look_ahead_size;
-        pool_size += look_ahead_size;
-
-    #ifdef DEBUG_SYCL_MALLOC
-        fprintf(stderr, "%s[%d]: %d buffers, max_size = %u MB, pool_size = %u MB, requested %u MB\n", __func__, id, nnz,
-                (uint32_t)(max_size/1024/1024), (uint32_t)(g_sycl_pool_size[id]/1024/1024), (uint32_t)(size/1024/1024));
-    #endif
-        // GGML_SYCL_DEBUG("ggml_sycl_pool_malloc_leg look_ahead_size=%lu, return %p\n", look_ahead_size, ptr);
-        return ptr;
-    }
-
-    void free(void * ptr, size_t size) override {
-        for (int i = 0; i < MAX_SYCL_BUFFERS; ++i) {
-            ggml_sycl_buffer& b = buffer_pool[i];
-            if (b.ptr == nullptr) {
-                b.ptr = ptr;
-                b.size = size;
-                return;
-            }
-        }
-        fprintf(stderr, "WARNING: sycl buffer pool full, increase MAX_sycl_BUFFERS\n");
-        SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(ptr, *qptr)));
-        pool_size -= size;
-    }
-};
-
-std::unique_ptr ggml_backend_sycl_context::new_pool_for_device(queue_ptr qptr, int device) {
-    // TBD: NO VMM support
-    // if (ggml_sycl_info().devices[device].vmm) {
-    //     return std::unique_ptr(new ggml_sycl_pool_vmm(device));
-    // }
-   return std::unique_ptr(new ggml_sycl_pool_leg(qptr, device));
-}
-
-// TBD pool with virtual memory management
-// struct ggml_sycl_pool_vmm : public ggml_sycl_pool
-
 static dpct::err0 ggml_sycl_cpy_tensor_2d(void *dst,
                                           const struct ggml_tensor *src,
                                           int64_t i3, int64_t i2,
@@ -2376,54 +3229,6 @@ inline void ggml_sycl_op_pad(ggml_backend_sycl_context & ctx, const ggml_tensor
     (void) src1_dd;
 }
 
-static int64_t get_row_rounding(ggml_type type, const std::array & tensor_split) {
-    int64_t min_compute_capability = INT_MAX;
-    int64_t max_compute_capability = INT_MIN;
-    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
-        if (tensor_split[i] < (i + 1 < ggml_sycl_info().device_count ? tensor_split[i + 1] : 1.0f)) {
-            if (min_compute_capability > ggml_sycl_info().devices[i].cc) {
-                min_compute_capability = ggml_sycl_info().devices[i].cc;
-            }
-            if (max_compute_capability < ggml_sycl_info().devices[i].cc) {
-                max_compute_capability = ggml_sycl_info().devices[i].cc;
-            }
-        }
-    }
-
-    switch(type) {
-        case GGML_TYPE_Q4_0:
-        case GGML_TYPE_Q4_1:
-            return max_compute_capability >= VER_GEN9 ? 128 : 64;
-        case GGML_TYPE_Q5_0:
-        case GGML_TYPE_Q5_1:
-        case GGML_TYPE_Q8_0:
-            return 64;
-        case GGML_TYPE_F16:
-        case GGML_TYPE_F32:
-            return 1;
-        case GGML_TYPE_Q2_K:
-        case GGML_TYPE_Q3_K:
-        case GGML_TYPE_Q4_K:
-        case GGML_TYPE_Q5_K:
-        case GGML_TYPE_IQ2_XXS:
-        case GGML_TYPE_IQ2_XS:
-        case GGML_TYPE_IQ2_S:
-        case GGML_TYPE_IQ1_S:
-        case GGML_TYPE_IQ1_M:
-        case GGML_TYPE_IQ3_XXS:
-        case GGML_TYPE_IQ4_XS:
-        case GGML_TYPE_IQ4_NL:
-            return max_compute_capability >= VER_GEN9 ? 128 : 64;
-        case GGML_TYPE_IQ3_S:
-            return max_compute_capability >= VER_GEN9 ? 128 : 64;
-        case GGML_TYPE_Q6_K:
-            return 64;
-        default:
-            GGML_ABORT("fatal error");
-    }
-
-}
-
 inline void ggml_sycl_op_mul_mat_sycl(
     ggml_backend_sycl_context & ctx,
     const ggml_tensor *src0, const ggml_tensor *src1, ggml_tensor *dst,
@@ -2783,10 +3588,6 @@ static void ggml_sycl_set_peer_access(const int n_tokens, int main_device) {
     peer_access_enabled = enable_peer_access;
 }
 
-struct ggml_backend_sycl_split_buffer_type_context {
-    std::array tensor_split;
-};
-
 static void ggml_sycl_op_mul_mat(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
                                  const ggml_tensor *src1, ggml_tensor *dst,
                                  ggml_sycl_op_mul_mat_t op,
@@ -3865,12 +4666,6 @@ static void ggml_sycl_nop(ggml_backend_sycl_context & ctx, const ggml_tensor * s
     (void) dst;
 }
 
-static size_t ggml_nbytes_split(const struct ggml_tensor * tensor, int nrows_split) {
-    static_assert(GGML_MAX_DIMS == 4, "GGML_MAX_DIMS is not 4 - update this function");
-
-    return nrows_split*ggml_row_size(tensor->type, tensor->ne[0]);
-}
-
 void ggml_sycl_set_main_device(const int main_device) try {
     if (dpct::get_current_device_id() == main_device) return;
     check_allow_gpu_index(main_device);
@@ -4038,39 +4833,9 @@ bool ggml_sycl_compute_forward(ggml_backend_sycl_context & ctx, struct ggml_tens
     return true;
 }
 
-GGML_API void   ggml_sycl_get_gpu_list(int *id_list, int max_len) try {
-    GGML_SYCL_DEBUG("[SYCL] call ggml_sycl_get_gpu_list\n");
-    for(int i=0;i=max_len) break;
-        id_list[i] = i;
-    }
-    return;
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-int ggml_sycl_get_device_count() try {
-    int device_count;
-    if (CHECK_TRY_ERROR(device_count =
-                             dpct::dev_mgr::instance().device_count()) != 0) {
-        return 0;
-    }
-    return device_count;
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-GGML_API void ggml_sycl_get_device_description(int device, char *description,
+GGML_API void ggml_backend_sycl_get_device_description(int device, char *description,
                                       size_t description_size) try {
-    GGML_SYCL_DEBUG("[SYCL] call ggml_sycl_get_device_description\n");
+    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_get_device_description\n");
     dpct::device_info prop;
     SYCL_CHECK(CHECK_TRY_ERROR(dpct::get_device_info(
         prop, dpct::dev_mgr::instance().get_device(device))));
@@ -4108,801 +4873,9 @@ catch (sycl::exception const &exc) {
 
 ////////////////////////////////////////////////////////////////////////////////
 
-// backend interface
-
-#define UNUSED GGML_UNUSED
-
-// sycl buffer
-
-struct ggml_backend_sycl_buffer_context {
-    int device;
-    void * dev_ptr = nullptr;
-    queue_ptr stream;
-    std::string name;
-
-     ggml_backend_sycl_buffer_context(int device, void * dev_ptr, queue_ptr stream) :
-        device(device), dev_ptr(dev_ptr), stream(stream) {
-            check_allow_gpu_index(device);
-            name = (GGML_SYCL_NAME + std::to_string(device));
-        }
-
-
-    ~ggml_backend_sycl_buffer_context() {
-        if (dev_ptr != nullptr) {
-            ggml_sycl_set_device(device);
-            SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(dev_ptr, *stream)));
-        }
-    }
-};
-
-static const char * ggml_backend_sycl_buffer_get_name(ggml_backend_buffer_t buffer) {
-    ggml_backend_sycl_buffer_context * ctx = (ggml_backend_sycl_buffer_context *)buffer->context;
-    return ctx->name.c_str();
-}
-
-static bool ggml_backend_buffer_is_sycl(ggml_backend_buffer_t buffer) {
-    return buffer->iface.get_name == ggml_backend_sycl_buffer_get_name;
-}
-
-static void
-ggml_backend_sycl_buffer_free_buffer(ggml_backend_buffer_t buffer) try {
-    ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
-    ggml_sycl_set_device(ctx->device);
-
-    delete ctx;
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-static void * ggml_backend_sycl_buffer_get_base(ggml_backend_buffer_t buffer) {
-    ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
-    return ctx->dev_ptr;
-}
-
-static void
-ggml_backend_sycl_buffer_init_tensor(ggml_backend_buffer_t buffer,
-                                     ggml_tensor *tensor) try {
-    ggml_backend_sycl_buffer_context * ctx = (ggml_backend_sycl_buffer_context *)buffer->context;
-
-    if (tensor->view_src != NULL && tensor->view_offs == 0) {
-        assert(tensor->view_src->buffer->buft == buffer->buft);
-        tensor->backend = tensor->view_src->backend;
-        tensor->extra = tensor->view_src->extra;
-        return;
-    }
-
-
-    if (ggml_is_quantized(tensor->type)) {
-        // initialize padding to 0 to avoid possible NaN values
-        size_t original_size = ggml_nbytes(tensor);
-        size_t padded_size = ggml_backend_buft_get_alloc_size(buffer->buft, tensor);
-
-        if (padded_size > original_size && tensor->view_src == nullptr) {
-            SYCL_CHECK(CHECK_TRY_ERROR(ctx->stream->memset(
-                (char *)tensor->data + original_size, 0,
-                padded_size - original_size).wait()));
-        }
-    }
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-static void ggml_backend_sycl_buffer_set_tensor(ggml_backend_buffer_t buffer,
-                                                ggml_tensor *tensor,
-                                                const void *data, size_t offset,
-                                                size_t size) try {
-
-    ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
-
-    ggml_sycl_set_device(ctx->device);
-    auto stream = &(dpct::dev_mgr::instance().get_device(ctx->device).default_queue());
-    SYCL_CHECK(
-        CHECK_TRY_ERROR(dpct::dev_mgr::instance().get_device(ctx->device).queues_wait_and_throw()));
-    char* host_buf = (char*)malloc(size);
-    memcpy(host_buf, data, size);
-    SYCL_CHECK(
-        CHECK_TRY_ERROR((*stream).memcpy((char *)tensor->data + offset, host_buf, size)
-                             .wait()));
-    free(host_buf);
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-static void ggml_backend_sycl_buffer_get_tensor(ggml_backend_buffer_t buffer,
-                                                const ggml_tensor *tensor,
-                                                void *data, size_t offset,
-                                                size_t size) try {
-
-    ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
-
-    ggml_sycl_set_device(ctx->device);
-    auto stream = dpct::dev_mgr::instance().get_device(ctx->device).default_queue();
-
-    SYCL_CHECK(CHECK_TRY_ERROR(
-        stream.memcpy(data, (const char *)tensor->data + offset, size)
-            .wait()));
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-static bool
-ggml_backend_sycl_buffer_cpy_tensor(ggml_backend_buffer_t buffer,
-                                    const ggml_tensor *src,
-                                    ggml_tensor *dst) try {
-    if (ggml_backend_buffer_is_sycl(src->buffer)) {
-        ggml_backend_sycl_buffer_context * src_ctx = (ggml_backend_sycl_buffer_context *)src->buffer->context;
-        ggml_backend_sycl_buffer_context * dst_ctx = (ggml_backend_sycl_buffer_context *)dst->buffer->context;
-
-        ggml_sycl_set_device(src_ctx->device);
-        /*
-        DPCT1009:198: SYCL uses exceptions to report errors and does not use the
-        error codes. The original code was commented out and a warning string
-        was inserted. You need to rewrite this code.
-        */
-        SYCL_CHECK(CHECK_TRY_ERROR(
-            dpct::dev_mgr::instance().get_device(src_ctx->device).queues_wait_and_throw()));
-        ggml_sycl_set_device(dst_ctx->device);
-        /*
-        DPCT1009:199: SYCL uses exceptions to report errors and does not use the
-        error codes. The original code was commented out and a warning string
-        was inserted. You need to rewrite this code.
-        */
-        SYCL_CHECK(CHECK_TRY_ERROR(
-            dpct::dev_mgr::instance().get_device(dst_ctx->device).queues_wait_and_throw()));
-        /*
-        DPCT1009:200: SYCL uses exceptions to report errors and does not use the
-        error codes. The original code was commented out and a warning string
-        was inserted. You need to rewrite this code.
-        */
-
-        queue_ptr stream_dst = dst_ctx->stream;
-        queue_ptr stream_src = src_ctx->stream;
-        size_t size = ggml_nbytes(src);
-
-        //todo. it's dirty solutino to walkaroud known issue:device2device cross GPUs.
-        dev2dev_memcpy(*stream_dst, *stream_src, dst->data, src->data, size);
-
-//todo, it's known issue：error in device2device cross GPUs. reused when the issue is fixed. DON"T remove
-#if 0
-        SYCL_CHECK(CHECK_TRY_ERROR((*stream).memcpy(
-            (char *)dst->data, (const char *)src->data, size).wait()));
-
-        /*
-        DPCT1009:201: SYCL uses exceptions to report errors and does not use the
-        error codes. The original code was commented out and a warning string
-        was inserted. You need to rewrite this code.
-        */
-        SYCL_CHECK(CHECK_TRY_ERROR(
-            dpct::dev_mgr::instance().get_device(dst_ctx->device).queues_wait_and_throw()));
-#endif
-        return true;
-    }
-    return false;
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-
-static void ggml_backend_sycl_buffer_clear(ggml_backend_buffer_t buffer,
-                                           uint8_t value) try {
-     ggml_backend_sycl_buffer_context * ctx = ( ggml_backend_sycl_buffer_context *)buffer->context;
-
-    ggml_sycl_set_device(ctx->device);
-    queue_ptr stream = ctx->stream;
-    SYCL_CHECK(
-        CHECK_TRY_ERROR(dpct::get_current_device().queues_wait_and_throw()));
-
-    SYCL_CHECK(CHECK_TRY_ERROR((*stream)
-                                    .memset(ctx->dev_ptr, value, buffer->size)
-                                    .wait()));
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-static struct ggml_backend_buffer_i ggml_backend_sycl_buffer_interface = {
-    /* .get_name        = */ ggml_backend_sycl_buffer_get_name,
-    /* .free_buffer     = */ ggml_backend_sycl_buffer_free_buffer,
-    /* .get_base        = */ ggml_backend_sycl_buffer_get_base,
-    /* .init_tensor     = */ ggml_backend_sycl_buffer_init_tensor,
-    /* .memset_tensor   = */ NULL,
-    /* .set_tensor      = */ ggml_backend_sycl_buffer_set_tensor,
-    /* .get_tensor      = */ ggml_backend_sycl_buffer_get_tensor,
-    /* .cpy_tensor      = */ ggml_backend_sycl_buffer_cpy_tensor,
-    /* .clear           = */ ggml_backend_sycl_buffer_clear,
-    /* .reset           = */ NULL,
-};
-
-// sycl buffer type
-struct ggml_backend_sycl_buffer_type_context {
-    int device;
-    std::string name;
-
-    // each buffer type has its own stream
-    queue_ptr stream = nullptr;
-};
-
-static const char * ggml_backend_sycl_buffer_type_name(ggml_backend_buffer_type_t buft) {
-    ggml_backend_sycl_buffer_type_context * ctx = (ggml_backend_sycl_buffer_type_context *)buft->context;
-
-    return ctx->name.c_str();
-}
-static ggml_backend_buffer_t
-ggml_backend_sycl_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft,
-                                           size_t size) try {
-    ggml_backend_sycl_buffer_type_context * buft_ctx = (ggml_backend_sycl_buffer_type_context *)buft->context;
-    ggml_sycl_set_device(buft_ctx->device);
-    const queue_ptr stream = buft_ctx->stream;
-    size = std::max(size, (size_t)1); // syclMalloc returns null for size 0
-
-    void * dev_ptr;
-    SYCL_CHECK(CHECK_TRY_ERROR(dev_ptr = (void *)sycl::malloc_device(
-                                    size, *stream)));
-    if (!dev_ptr) {
-        fprintf(stderr, "%s: can't malloc %lu Bytes memory on device", __func__, size);
-        return nullptr;
-    }
-    ggml_backend_sycl_buffer_context * ctx = new  ggml_backend_sycl_buffer_context(buft_ctx->device, dev_ptr, buft_ctx->stream);
-    return ggml_backend_buffer_init(buft, ggml_backend_sycl_buffer_interface, ctx, size);
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-static size_t ggml_backend_sycl_buffer_type_get_alignment(ggml_backend_buffer_type_t buft) {
-    return 128;
-    UNUSED(buft);
-}
-
-static size_t ggml_backend_sycl_buffer_type_get_max_size(ggml_backend_buffer_type_t buft) {
-    return dpct::get_current_device().get_max_mem_alloc_size();
-
-    UNUSED(buft);
-}
-
-static size_t ggml_backend_sycl_buffer_type_get_alloc_size(ggml_backend_buffer_type_t buft, const ggml_tensor * tensor) {
-    size_t size = ggml_nbytes(tensor);
-    int64_t ne0 = tensor->ne[0];
-
-    if (ggml_is_quantized(tensor->type)) {
-        if (ne0 % MATRIX_ROW_PADDING != 0) {
-            size += ggml_row_size(tensor->type, MATRIX_ROW_PADDING - ne0 % MATRIX_ROW_PADDING);
-        }
-    }
-
-    return size;
-
-    UNUSED(buft);
-}
-
-static ggml_backend_buffer_type_i ggml_backend_sycl_buffer_type_interface = {
-    /* .get_name         = */ ggml_backend_sycl_buffer_type_name,
-    /* .alloc_buffer     = */ ggml_backend_sycl_buffer_type_alloc_buffer,
-    /* .get_alignment    = */ ggml_backend_sycl_buffer_type_get_alignment,
-    /* .get_max_size     = */ ggml_backend_sycl_buffer_type_get_max_size,
-    /* .get_alloc_size   = */ ggml_backend_sycl_buffer_type_get_alloc_size,
-    /* .is_host          = */ nullptr,
-};
-
-ggml_backend_buffer_type_t ggml_backend_sycl_buffer_type(int device) {
-    static std::mutex mutex;
-    std::lock_guard lock(mutex);
-
-    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_buffer_type\n");
-
-    if (device>=ggml_sycl_info().device_count or device<0) {
-        printf("ggml_backend_sycl_buffer_type error: device_index:%d is out of range [0, %d], miss to call ggml_backend_sycl_set_single_device()\n",
-            device, ggml_sycl_info().device_count-1);
-        GGML_ASSERT(devicedevice;
-    if (device>=ggml_sycl_info().device_count or device<0) {
-        printf("ggml_backend_sycl_buffer_type error: device_index:%d is out of range [0, %d], miss to call ggml_backend_sycl_set_single_device()\n",
-            device, ggml_sycl_info().device_count-1);
-        GGML_ASSERT(devicestream(i, 0)},
-            };
-        }
-        ggml_backend_sycl_buffer_type_initialized = true;
-    }
-    return &ggml_backend_sycl_buffer_types[device];
-}
-
-// sycl split buffer type
-static void get_row_split(int64_t * row_low, int64_t * row_high, const ggml_tensor * tensor, const std::array & tensor_split, int id) {
-    const int64_t nrows = ggml_nrows(tensor);
-    const int64_t rounding = get_row_rounding(tensor->type, tensor_split);
-
-    *row_low = id == 0 ? 0 : nrows*tensor_split[id];
-    *row_low -= *row_low % rounding;
-    if (id == ggml_sycl_info().device_count - 1) {
-        *row_high = nrows;
-    } else {
-        *row_high = nrows*tensor_split[id + 1];
-        *row_high -= *row_high % rounding;
-    }
-}
-
-struct ggml_backend_sycl_split_buffer_context {
-    ~ggml_backend_sycl_split_buffer_context() try {
-        for (ggml_tensor_extra_gpu * extra : tensor_extras) {
-            for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
-                for (int64_t is = 0; is < GGML_SYCL_MAX_STREAMS; ++is) {
-                    if (extra->events[i][is] != nullptr) {
-                        /*
-                        DPCT1009:206: SYCL uses exceptions to report errors and
-                        does not use the error codes. The original code was
-                        commented out and a warning string was inserted. You
-                        need to rewrite this code.
-                        */
-                        SYCL_CHECK(CHECK_TRY_ERROR(
-                            dpct::destroy_event(extra->events[i][is])));
-                    }
-                }
-                if (extra->data_device[i] != nullptr) {
-                    /*
-                    DPCT1009:207: SYCL uses exceptions to report errors and does
-                    not use the error codes. The original code was commented out
-                    and a warning string was inserted. You need to rewrite this
-                    code.
-                    */
-                    ggml_sycl_set_device(i);
-                    SYCL_CHECK(CHECK_TRY_ERROR(sycl::free(
-                        extra->data_device[i], *(streams[i]))));
-                }
-            }
-            delete extra;
-        }
-    }
-    catch (sycl::exception const &exc) {
-      std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-                << ", line:" << __LINE__ << std::endl;
-      std::exit(1);
-    }
-
-    std::vector tensor_extras;
-    std::vector streams;
-};
-
-static const char * ggml_backend_sycl_split_buffer_get_name(ggml_backend_buffer_t buffer) {
-    return GGML_SYCL_NAME "_Split";
-
-    UNUSED(buffer);
-}
-
-static bool ggml_backend_buffer_is_sycl_split(ggml_backend_buffer_t buffer) {
-   return buffer->iface.get_name == ggml_backend_sycl_split_buffer_get_name;
-}
-
-static void ggml_backend_sycl_split_buffer_free_buffer(ggml_backend_buffer_t buffer) {
-    ggml_backend_sycl_split_buffer_context * ctx = (ggml_backend_sycl_split_buffer_context *)buffer->context;
-    delete ctx;
-}
-
-static void * ggml_backend_sycl_split_buffer_get_base(ggml_backend_buffer_t buffer) {
-    // the pointers are stored in the tensor extras, this is just a dummy address and never dereferenced
-    return (void *)0x1000;
-
-    UNUSED(buffer);
-}
-
-static void
-ggml_backend_sycl_split_buffer_init_tensor(ggml_backend_buffer_t buffer,
-                                           ggml_tensor *tensor) try {
-    GGML_ASSERT(tensor->view_src == nullptr); // views of split tensors are not supported
-
-    ggml_backend_sycl_split_buffer_context * ctx = (ggml_backend_sycl_split_buffer_context *)buffer->context;
-    ggml_backend_sycl_split_buffer_type_context * buft_ctx = (ggml_backend_sycl_split_buffer_type_context *)buffer->buft->context;
-
-    const int64_t ne0 = tensor->ne[0];
-
-    ggml_tensor_extra_gpu * extra = new ggml_tensor_extra_gpu{};
-
-    ctx->tensor_extras.push_back(extra);
-        ctx->streams.push_back(&(dpct::get_current_device().default_queue()));
-
-    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
-        int64_t row_low, row_high;
-        get_row_split(&row_low, &row_high, tensor, buft_ctx->tensor_split, i);
-
-        int64_t nrows_split = row_high - row_low;
-        if (nrows_split == 0) {
-            continue;
-        }
-
-        size_t size = ggml_nbytes_split(tensor, nrows_split);
-        const size_t original_size = size;
-
-        // pad last row to a multiple of 512 elements to avoid out-of-bounds memory accesses
-        if (ne0 % MATRIX_ROW_PADDING != 0) {
-            size += ggml_row_size(tensor->type, MATRIX_ROW_PADDING - ne0 % MATRIX_ROW_PADDING);
-        }
-
-        // FIXME: do not crash if cudaMalloc fails
-        // currently, init_tensor cannot fail, it needs to be fixed in ggml-backend first
-        ggml_sycl_set_device(i);
-        const queue_ptr stream = ctx->streams[i];
-        char * buf;
-        /*
-        DPCT1009:208: SYCL uses exceptions to report errors and does not use the
-        error codes. The original code was commented out and a warning string
-        was inserted. You need to rewrite this code.
-        */
-        SYCL_CHECK(CHECK_TRY_ERROR(buf = (char *)sycl::malloc_device(
-                                        size, *stream)));
-        if (!buf) {
-            char err_buf[1024];
-            snprintf(err_buf, 1023, "%s: can't malloc %lu Bytes memory on device", __func__, size);
-            throw std::runtime_error(err_buf);
-        }
-        // set padding to 0 to avoid possible NaN values
-        if (size > original_size) {
-            /*
-            DPCT1009:209: SYCL uses exceptions to report errors and does not use
-            the error codes. The original code was commented out and a warning
-            string was inserted. You need to rewrite this code.
-            */
-            SYCL_CHECK(CHECK_TRY_ERROR(
-                (*stream)
-                    .memset(buf + original_size, 0, size - original_size)
-                    .wait()));
-        }
-
-        extra->data_device[i] = buf;
-
-        for (int64_t is = 0; is < GGML_SYCL_MAX_STREAMS; ++is) {
-            /*
-            DPCT1009:210: SYCL uses exceptions to report errors and does not use
-            the error codes. The original code was commented out and a warning
-            string was inserted. You need to rewrite this code.
-            */
-            SYCL_CHECK(
-                CHECK_TRY_ERROR(extra->events[i][is] = new sycl::event()));
-        }
-    }
-    tensor->backend = GGML_BACKEND_TYPE_GPU_SPLIT;
-    tensor->extra = extra;
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-static void
-ggml_backend_sycl_split_buffer_set_tensor(ggml_backend_buffer_t buffer,
-                                          ggml_tensor *tensor, const void *data,
-                                          size_t offset, size_t size) try {
-    // split tensors must always be set in their entirety at once
-    GGML_ASSERT(offset == 0);
-    GGML_ASSERT(size == ggml_nbytes(tensor));
-
-    ggml_backend_sycl_split_buffer_context * ctx = (ggml_backend_sycl_split_buffer_context *)buffer->context;
-    ggml_backend_sycl_split_buffer_type_context * buft_ctx = (ggml_backend_sycl_split_buffer_type_context *)buffer->buft->context;
-
-    const int64_t ne0 = tensor->ne[0];
-    const size_t nb1 = tensor->nb[1];
-    ggml_tensor_extra_gpu * extra = (ggml_tensor_extra_gpu *)tensor->extra;
-
-    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
-        int64_t row_low, row_high;
-        get_row_split(&row_low, &row_high, tensor, buft_ctx->tensor_split, i);
-
-        int64_t nrows_split = row_high - row_low;
-        if (nrows_split == 0) {
-            continue;
-        }
-
-        const size_t offset_split = row_low*nb1;
-        size_t size = ggml_nbytes_split(tensor, nrows_split);
-        const size_t original_size = size;
-
-        // pad last row to a multiple of 512 elements to avoid out-of-bounds memory accesses
-        if (ne0 % MATRIX_ROW_PADDING != 0) {
-            size += ggml_row_size(tensor->type, MATRIX_ROW_PADDING - ne0 % MATRIX_ROW_PADDING);
-        }
-
-        const char * buf_host = (const char *)data + offset_split;
-        /*
-        DPCT1009:211: SYCL uses exceptions to report errors and does not use the
-        error codes. The original code was commented out and a warning string
-        was inserted. You need to rewrite this code.
-        */
-        ggml_sycl_set_device(i);
-        const queue_ptr stream = ctx->streams[i];
-        SYCL_CHECK(CHECK_TRY_ERROR(
-            (*stream)
-                .memcpy(extra->data_device[i], buf_host, original_size)
-                .wait()));
-    }
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-static void
-ggml_backend_sycl_split_buffer_get_tensor(ggml_backend_buffer_t buffer,
-                                          const ggml_tensor *tensor, void *data,
-                                          size_t offset, size_t size) try {
-    // split tensors must always be set in their entirety at once
-    GGML_ASSERT(offset == 0);
-    GGML_ASSERT(size == ggml_nbytes(tensor));
-
-    ggml_backend_sycl_split_buffer_context * ctx = (ggml_backend_sycl_split_buffer_context *)buffer->context;
-    ggml_backend_sycl_split_buffer_type_context * buft_ctx = (ggml_backend_sycl_split_buffer_type_context *)buffer->buft->context;
-
-    const int64_t ne0 = tensor->ne[0];
-    const size_t nb1 = tensor->nb[1];
-    ggml_tensor_extra_gpu * extra = (ggml_tensor_extra_gpu *)tensor->extra;
-
-    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
-        int64_t row_low, row_high;
-        get_row_split(&row_low, &row_high, tensor, buft_ctx->tensor_split, i);
-
-        int64_t nrows_split = row_high - row_low;
-        if (nrows_split == 0) {
-            continue;
-        }
-
-        const size_t offset_split = row_low*nb1;
-        size_t size = ggml_nbytes_split(tensor, nrows_split);
-        const size_t original_size = size;
-
-        // pad last row to a multiple of 512 elements to avoid out-of-bounds memory accesses
-        if (ne0 % MATRIX_ROW_PADDING != 0) {
-            size += ggml_row_size(tensor->type, MATRIX_ROW_PADDING - ne0 % MATRIX_ROW_PADDING);
-        }
-
-        char * buf_host = (char *)data + offset_split;
-        /*
-        DPCT1009:212: SYCL uses exceptions to report errors and does not use the
-        error codes. The original code was commented out and a warning string
-        was inserted. You need to rewrite this code.
-        */
-        ggml_sycl_set_device(i);
-        const queue_ptr stream = ctx->streams[i];
-        SYCL_CHECK(CHECK_TRY_ERROR(
-            (*stream)
-                .memcpy(buf_host, extra->data_device[i], original_size)
-                .wait()));
-    }
-}
-catch (sycl::exception const &exc) {
-  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
-            << ", line:" << __LINE__ << std::endl;
-  std::exit(1);
-}
-
-static void ggml_backend_sycl_split_buffer_clear(ggml_backend_buffer_t buffer, uint8_t value) {
-    UNUSED(buffer);
-    UNUSED(value);
-}
-
-static struct ggml_backend_buffer_i ggml_backend_sycl_split_buffer_interface = {
-    /* .get_name        = */ ggml_backend_sycl_split_buffer_get_name,
-    /* .free_buffer     = */ ggml_backend_sycl_split_buffer_free_buffer,
-    /* .get_base        = */ ggml_backend_sycl_split_buffer_get_base,
-    /* .init_tensor     = */ ggml_backend_sycl_split_buffer_init_tensor,
-    /* .memset_tensor   = */ NULL,
-    /* .set_tensor      = */ ggml_backend_sycl_split_buffer_set_tensor,
-    /* .get_tensor      = */ ggml_backend_sycl_split_buffer_get_tensor,
-    /* .cpy_tensor      = */ NULL,
-    /* .clear           = */ ggml_backend_sycl_split_buffer_clear,
-    /* .reset           = */ NULL,
-};
-
-static const char * ggml_backend_sycl_split_buffer_type_name(ggml_backend_buffer_type_t buft) {
-    return GGML_SYCL_NAME "_Split";
-
-    UNUSED(buft);
-}
-
-static ggml_backend_buffer_t ggml_backend_sycl_split_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
-    // since we don't know the exact split after rounding, we cannot allocate the device buffers at this point
-    // instead, we allocate them for each tensor separately in init_tensor
-    // however, the size still represents the maximum cumulative size of all the device buffers after the tensors are allocated,
-    // as returned by get_alloc_size. this limit is enforced during tensor allocation by ggml-alloc, so it must be correct.
-    ggml_backend_sycl_split_buffer_context * ctx = new ggml_backend_sycl_split_buffer_context();
-
-    return ggml_backend_buffer_init(buft, ggml_backend_sycl_split_buffer_interface, ctx, size);
-}
-
-static size_t ggml_backend_sycl_split_buffer_type_get_alignment(ggml_backend_buffer_type_t buft) {
-    return 128;
-    UNUSED(buft);
-}
-
-static size_t ggml_backend_sycl_split_buffer_type_get_alloc_size(ggml_backend_buffer_type_t buft, const ggml_tensor * tensor) {
-    ggml_backend_sycl_split_buffer_type_context * ctx = (ggml_backend_sycl_split_buffer_type_context *)buft->context;
-
-    size_t total_size = 0;
-
-    const int64_t ne0 = tensor->ne[0];
-
-    for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
-        int64_t row_low, row_high;
-        get_row_split(&row_low, &row_high, tensor, ctx->tensor_split, i);
-
-        int64_t nrows_split = row_high - row_low;
-        if (nrows_split == 0) {
-            continue;
-        }
-
-        total_size += ggml_nbytes_split(tensor, nrows_split);
-
-        // pad last row to a multiple of 512 elements to avoid out-of-bounds memory accesses
-        if (ne0 % MATRIX_ROW_PADDING != 0) {
-            total_size += ggml_row_size(tensor->type, MATRIX_ROW_PADDING - ne0 % MATRIX_ROW_PADDING);
-        }
-    }
-
-    return total_size;
-}
-
-static bool ggml_backend_sycl_split_buffer_type_is_host(ggml_backend_buffer_type_t buft) {
-    return false;
-
-    UNUSED(buft);
-}
-
-static ggml_backend_buffer_type_i ggml_backend_sycl_split_buffer_type_interface = {
-    /* .get_name         = */ ggml_backend_sycl_split_buffer_type_name,
-    /* .alloc_buffer     = */ ggml_backend_sycl_split_buffer_type_alloc_buffer,
-    /* .get_alignment    = */ ggml_backend_sycl_split_buffer_type_get_alignment,
-    /* .get_max_size     = */ NULL, // defaults to SIZE_MAX
-    /* .get_alloc_size   = */ ggml_backend_sycl_split_buffer_type_get_alloc_size,
-    /* .is_host          = */ ggml_backend_sycl_split_buffer_type_is_host,
-};
-
-ggml_backend_buffer_type_t ggml_backend_sycl_split_buffer_type(const float * tensor_split) {
-    static std::mutex mutex;
-    std::lock_guard lock(mutex);
-
-    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_split_buffer_type\n");
-    ggml_check_sycl();
-    // FIXME: this is not thread safe
-    static std::map, struct ggml_backend_buffer_type> buft_map;
-
-    std::array tensor_split_arr = {};
-
-    bool all_zero = tensor_split == nullptr || std::all_of(tensor_split, tensor_split + GGML_SYCL_MAX_DEVICES, [](float x) { return x == 0.0f; });
-    if (all_zero) {
-        tensor_split_arr = ggml_sycl_info().default_tensor_split;
-    } else {
-        float split_sum = 0.0f;
-        for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
-            tensor_split_arr[i] = split_sum;
-            split_sum += tensor_split[i];
-        }
-        for (int i = 0; i < ggml_sycl_info().device_count; ++i) {
-            tensor_split_arr[i] /= split_sum;
-        }
-    }
-
-    auto it = buft_map.find(tensor_split_arr);
-    if (it != buft_map.end()) {
-        return &it->second;
-    }
-
-    struct ggml_backend_buffer_type buft {
-        /* .iface   = */ ggml_backend_sycl_split_buffer_type_interface,
-        /* .device  = */ nullptr,
-        /* .context = */ new ggml_backend_sycl_split_buffer_type_context{tensor_split_arr},
-    };
-
-    auto result = buft_map.emplace(tensor_split_arr, buft);
-    return &result.first->second;
-}
-
-// host buffer type
-
-static const char * ggml_backend_sycl_host_buffer_type_name(ggml_backend_buffer_type_t buft) {
-    return GGML_SYCL_NAME "_Host";
-
-    UNUSED(buft);
-}
-
-static const char * ggml_backend_sycl_host_buffer_name(ggml_backend_buffer_t buffer) {
-    return GGML_SYCL_NAME "_Host";
-
-    UNUSED(buffer);
-}
-
-static void ggml_backend_sycl_host_buffer_free_buffer(ggml_backend_buffer_t buffer) {
-    ggml_sycl_host_free(buffer->context);
-}
-
-static ggml_backend_buffer_t ggml_backend_sycl_host_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
-    void * ptr = ggml_sycl_host_malloc(size);
-
-    if (ptr == nullptr) {
-        // fallback to cpu buffer
-        return ggml_backend_buft_alloc_buffer(ggml_backend_cpu_buffer_type(), size);
-    }
-
-    // FIXME: this is a hack to avoid having to implement a new buffer type
-    ggml_backend_buffer_t buffer = ggml_backend_cpu_buffer_from_ptr(ptr, size);
-    buffer->buft = buft;
-    buffer->iface.get_name = ggml_backend_sycl_host_buffer_name;
-    buffer->iface.free_buffer = ggml_backend_sycl_host_buffer_free_buffer;
-
-    return buffer;
-}
-
-ggml_backend_buffer_type_t ggml_backend_sycl_host_buffer_type() {
-    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_host_buffer_type\n");
-    static struct ggml_backend_buffer_type ggml_backend_sycl_buffer_type_host = {
-        /* .iface    = */ {
-            /* .get_name         = */ ggml_backend_sycl_host_buffer_type_name,
-            /* .alloc_buffer     = */ ggml_backend_sycl_host_buffer_type_alloc_buffer,
-            /* .get_alignment    = */ ggml_backend_cpu_buffer_type()->iface.get_alignment,
-            /* .get_max_size     = */ NULL, // TODO: return device.maxBufferLength
-            /* .get_alloc_size   = */ ggml_backend_cpu_buffer_type()->iface.get_alloc_size,
-            /* .is_host          = */ ggml_backend_cpu_buffer_type()->iface.is_host,
-        },
-        /* .device   = */ nullptr,
-        /* .context  = */ nullptr,
-    };
-
-    return &ggml_backend_sycl_buffer_type_host;
-}
-
 // backend
 
-static const char * ggml_backend_sycl_name(ggml_backend_t backend) {
+static const char * ggml_backend_sycl_get_name(ggml_backend_t backend) {
 
     ggml_backend_sycl_context * sycl_ctx = (ggml_backend_sycl_context *)backend->context;
 
@@ -4931,8 +4904,8 @@ static void ggml_backend_sycl_set_tensor_async(ggml_backend_t backend,
 
     GGML_ASSERT(buf->buft == ggml_backend_sycl_buffer_type(sycl_ctx->device) && "unsupported buffer type");
     const queue_ptr stream = sycl_ctx->stream(sycl_ctx->device, 0);
-    SYCL_CHECK(CHECK_TRY_ERROR((stream)->memcpy(
-        (char *)tensor->data + offset, data, size).wait()));
+    SYCL_CHECK(CHECK_TRY_ERROR(
+        (stream)->memcpy((char *)tensor->data + offset, data, size)));
 }
 catch (sycl::exception const &exc) {
   std::cerr << exc.what() << "Exception caught at file:" << __FILE__
@@ -4987,7 +4960,7 @@ static void ggml_backend_sycl_synchronize(ggml_backend_t backend) try {
     const queue_ptr stream = sycl_ctx->stream(sycl_ctx->device, 0);
     SYCL_CHECK(CHECK_TRY_ERROR((stream)->wait()));
 
-    UNUSED(backend);
+    GGML_UNUSED(backend);
 }
 catch (sycl::exception const &exc) {
   std::cerr << exc.what() << "Exception caught at file:" << __FILE__
@@ -5023,7 +4996,151 @@ static ggml_status ggml_backend_sycl_graph_compute(ggml_backend_t backend, ggml_
     return GGML_STATUS_SUCCESS;
 }
 
-static bool ggml_backend_sycl_supports_op(ggml_backend_t backend, const ggml_tensor * op) {
+static void ggml_backend_sycl_event_record(ggml_backend_t backend, ggml_backend_event_t event)
+try
+{
+    ggml_backend_sycl_context *sycl_ctx =
+        (ggml_backend_sycl_context *)backend->context;
+    sycl::event *sycl_event = static_cast(event->context);
+
+    const queue_ptr &stream = sycl_ctx->stream(sycl_ctx->device, 0);
+    // Record the current state of the queue
+    SYCL_CHECK(CHECK_TRY_ERROR(*sycl_event = stream->ext_oneapi_submit_barrier()));
+}
+catch (sycl::exception const &exc)
+{
+    std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+              << ", line:" << __LINE__ << std::endl;
+    std::exit(1);
+}
+
+static void ggml_backend_sycl_event_wait(ggml_backend_t backend, ggml_backend_event_t event) try {
+    ggml_backend_sycl_context* sycl_ctx = static_cast(backend->context);
+    sycl::event* sycl_event = static_cast(event->context);
+
+    if (ggml_backend_is_sycl(backend)) {
+        SYCL_CHECK(CHECK_TRY_ERROR(sycl_event->wait()));
+    } else
+        GGML_ABORT("fatal error");
+} catch (sycl::exception const& exc) {
+    std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+              << ", line:" << __LINE__ << std::endl;
+    std::exit(1);
+}
+
+static ggml_backend_i ggml_backend_sycl_interface = {
+    /* .get_name                = */ ggml_backend_sycl_get_name,
+    /* .free                    = */ ggml_backend_sycl_free,
+    /* .get_default_buffer_type = */ ggml_backend_sycl_get_default_buffer_type,
+    /* .set_tensor_async        = */ ggml_backend_sycl_set_tensor_async,
+    /* .get_tensor_async        = */ ggml_backend_sycl_get_tensor_async,
+    /* .cpy_tensor_async        = */ NULL, // ggml_backend_sycl_cpy_tensor_async,
+                                           // // TODO: update for the new
+                                           // interface
+    /* .synchronize             = */ ggml_backend_sycl_synchronize,
+    /* .graph_plan_create       = */ NULL,
+    /* .graph_plan_free         = */ NULL,
+    /* .graph_plan_update       = */ NULL,
+    /* .graph_plan_compute      = */ NULL,
+    /* .graph_compute           = */ ggml_backend_sycl_graph_compute,
+    /* .supports_op             = */ NULL, // moved to device
+    /* .supports_buft           = */ NULL, // moved to device
+    /* .offload_op              = */ NULL, // moved to device
+    /* .event_record            = */ ggml_backend_sycl_event_record,
+    /* .event_wait              = */ ggml_backend_sycl_event_wait,
+};
+
+static ggml_guid_t ggml_backend_sycl_guid() {
+    static ggml_guid guid = { 0x58, 0x05, 0x13, 0x8f, 0xcd, 0x3a, 0x61, 0x9d, 0xe7, 0xcd, 0x98, 0xa9, 0x03, 0xfd, 0x7c, 0x53 };
+    return &guid;
+}
+
+bool ggml_backend_is_sycl(ggml_backend_t backend) {
+    return backend != NULL && ggml_guid_matches(backend->guid, ggml_backend_sycl_guid());
+}
+
+int ggml_backend_sycl_get_device_count() {
+    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_get_device_count\n");
+    return ggml_sycl_info().device_count;
+}
+
+
+// backend device
+
+struct ggml_backend_sycl_device_context {
+    int device;
+    std::string name;
+    std::string description;
+};
+
+static const char * ggml_backend_sycl_device_get_name(ggml_backend_dev_t dev) {
+    ggml_backend_sycl_device_context * ctx = (ggml_backend_sycl_device_context *)dev->context;
+    return ctx->name.c_str();
+}
+
+static const char * ggml_backend_sycl_device_get_description(ggml_backend_dev_t dev) {
+    ggml_backend_sycl_device_context * ctx = (ggml_backend_sycl_device_context *)dev->context;
+    return ctx->description.c_str();
+}
+
+static void ggml_backend_sycl_device_get_memory(ggml_backend_dev_t dev, size_t * free, size_t * total) {
+    ggml_backend_sycl_device_context * ctx = (ggml_backend_sycl_device_context *)dev->context;
+    ggml_sycl_set_device(ctx->device);
+    SYCL_CHECK(CHECK_TRY_ERROR(
+    dpct::dev_mgr::instance().get_device(ctx->device).get_memory_info(*free, *total)));
+}
+
+static enum ggml_backend_dev_type ggml_backend_sycl_device_get_type(ggml_backend_dev_t dev) {
+    GGML_UNUSED(dev);
+    return GGML_BACKEND_DEVICE_TYPE_GPU_FULL;
+}
+
+static void ggml_backend_sycl_device_get_props(ggml_backend_dev_t dev, ggml_backend_dev_props * props) {
+    props->name        = ggml_backend_sycl_device_get_name(dev);
+    props->description = ggml_backend_sycl_device_get_description(dev);
+    props->type        = ggml_backend_sycl_device_get_type(dev);
+    ggml_backend_sycl_device_get_memory(dev, &props->memory_free, &props->memory_total);
+
+    bool host_buffer = getenv("GGML_SYCL_NO_PINNED") == nullptr;
+#ifdef GGML_SYCL_NO_PEER_COPY
+    bool events = false;
+#else
+    bool events = true;
+#endif
+
+    props->caps = {
+        /* .async                 = */ true,
+        /* .host_buffer           = */ host_buffer,
+        /* .buffer_from_host_ptr  = */ false,
+        /* .events                = */ events,
+    };
+}
+
+static ggml_backend_t ggml_backend_sycl_device_init(ggml_backend_dev_t dev, const char * params) {
+    GGML_UNUSED(params);
+    ggml_backend_sycl_device_context * ctx = (ggml_backend_sycl_device_context *)dev->context;
+    return ggml_backend_sycl_init(ctx->device);
+}
+
+static ggml_backend_buffer_type_t ggml_backend_sycl_device_get_buffer_type(ggml_backend_dev_t dev) {
+    ggml_backend_sycl_device_context * ctx = (ggml_backend_sycl_device_context *)dev->context;
+    return ggml_backend_sycl_buffer_type(ctx->device);
+}
+
+static ggml_backend_buffer_type_t ggml_backend_sycl_device_get_host_buffer_type(ggml_backend_dev_t dev) {
+    GGML_UNUSED(dev);
+    return ggml_backend_sycl_host_buffer_type();
+}
+
+static ggml_backend_buffer_t ggml_backend_sycl_device_buffer_from_host_ptr(ggml_backend_dev_t dev, void * ptr, size_t size, size_t max_tensor_size) {
+    GGML_UNUSED(dev);
+    GGML_UNUSED(ptr);
+    GGML_UNUSED(size);
+    GGML_UNUSED(max_tensor_size);
+    return nullptr;
+}
+
+static bool ggml_backend_sycl_device_supports_op(ggml_backend_dev_t dev, const ggml_tensor * op) {
     switch (op->op) {
         case GGML_OP_CONV_TRANSPOSE_1D:
             {
@@ -5167,47 +5284,173 @@ static bool ggml_backend_sycl_supports_op(ggml_backend_t backend, const ggml_ten
             return false;
     }
 
-    UNUSED(backend);
+    GGML_UNUSED(dev);
 }
 
-static bool ggml_backend_sycl_offload_op(ggml_backend_t backend, const ggml_tensor * op) {
-    const int min_batch_size = 32;
-    return op->ne[1] >= min_batch_size && op->op != GGML_OP_GET_ROWS && op->op != GGML_OP_MUL_MAT_ID;
-    GGML_UNUSED(backend);
-}
-
-static bool ggml_backend_sycl_supports_buft(ggml_backend_t backend, ggml_backend_buffer_type_t buft) {
-    if (buft->iface.get_name != ggml_backend_sycl_buffer_type_name) {
+static bool ggml_backend_sycl_device_supports_buft(ggml_backend_dev_t dev, ggml_backend_buffer_type_t buft) {
+    if (buft->iface.get_name != ggml_backend_sycl_buffer_type_get_name) {
         return false;
     }
     ggml_backend_sycl_buffer_type_context * buft_ctx = (ggml_backend_sycl_buffer_type_context *)buft->context;
-    ggml_backend_sycl_context * sycl_ctx = (ggml_backend_sycl_context *)backend->context;
+    ggml_backend_sycl_device_context * sycl_ctx = (ggml_backend_sycl_device_context *)dev->context;
     return buft_ctx->device == sycl_ctx->device;
 }
 
-static ggml_backend_i ggml_backend_sycl_interface = {
-    /* .get_name                = */ ggml_backend_sycl_name,
-    /* .free                    = */ ggml_backend_sycl_free,
-    /* .get_default_buffer_type = */ ggml_backend_sycl_get_default_buffer_type,
-    /* .set_tensor_async        = */ ggml_backend_sycl_set_tensor_async,
-    /* .get_tensor_async        = */ ggml_backend_sycl_get_tensor_async,
-    /* .cpy_tensor_async        = */ NULL, //ggml_backend_sycl_cpy_tensor_async, // TODO: update for the new interface
-    /* .synchronize             = */ ggml_backend_sycl_synchronize,
-    /* .graph_plan_create       = */ NULL,
-    /* .graph_plan_free         = */ NULL,
-    /* .graph_plan_update       = */ NULL,
-    /* .graph_plan_compute      = */ NULL,
-    /* .graph_compute           = */ ggml_backend_sycl_graph_compute,
-    /* .supports_op             = */ ggml_backend_sycl_supports_op,
-    /* .supports_buft           = */ ggml_backend_sycl_supports_buft,
-    /* .offload_op              = */ ggml_backend_sycl_offload_op,
-    /* .event_record            = */ NULL,
-    /* .event_wait              = */ NULL,
+static bool ggml_backend_sycl_device_offload_op(ggml_backend_dev_t dev, const ggml_tensor * op) {
+    const int min_batch_size = 32;
+    return op->ne[1] >= min_batch_size && op->op != GGML_OP_GET_ROWS && op->op != GGML_OP_MUL_MAT_ID;
+    GGML_UNUSED(dev);
+}
+
+static ggml_backend_event_t
+ggml_backend_sycl_device_event_new(ggml_backend_dev_t dev) {
+
+#ifdef GGML_SYCL_NO_PEER_COPY
+    return nullptr;
+#else
+  sycl::event *event_ptr = new sycl::event();
+
+  return new ggml_backend_event{
+      /* .device = */ dev,
+      /* .context = */ event_ptr,
+  };
+#endif
+}
+
+static void ggml_backend_sycl_device_event_free(ggml_backend_dev_t dev, ggml_backend_event_t event) try {
+  GGML_UNUSED(dev);
+  if (event == nullptr) {
+    return;
+  }
+
+  if (event->context != nullptr) {
+    sycl::event *sycl_event = static_cast(event->context);
+    delete sycl_event;
+    event->context = nullptr;
+  }
+
+  delete event;
+} catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+
+static void ggml_backend_sycl_device_event_synchronize(ggml_backend_dev_t dev, ggml_backend_event_t event) try {
+  GGML_UNUSED(dev);
+
+  sycl::event *sycl_event = static_cast(event->context);
+  SYCL_CHECK(CHECK_TRY_ERROR(sycl_event->wait()));
+} catch (sycl::exception const &exc) {
+  std::cerr << exc.what() << "Exception caught at file:" << __FILE__
+            << ", line:" << __LINE__ << std::endl;
+  std::exit(1);
+}
+
+static const ggml_backend_device_i ggml_backend_sycl_device_interface = {
+    /* .get_name                = */ ggml_backend_sycl_device_get_name,
+    /* .get_description         = */ ggml_backend_sycl_device_get_description,
+    /* .get_memory              = */ ggml_backend_sycl_device_get_memory,
+    /* .get_type                = */ ggml_backend_sycl_device_get_type,
+    /* .get_props               = */ ggml_backend_sycl_device_get_props,
+    /* .init_backend            = */ ggml_backend_sycl_device_init,
+    /* .get_buffer_type         = */ ggml_backend_sycl_device_get_buffer_type,
+    /* .get_host_buffer_type    = */ ggml_backend_sycl_device_get_host_buffer_type,
+    /* .buffer_from_host_ptr    = */ ggml_backend_sycl_device_buffer_from_host_ptr,
+    /* .supports_op             = */ ggml_backend_sycl_device_supports_op,
+    /* .supports_buft           = */ ggml_backend_sycl_device_supports_buft,
+    /* .offload_op              = */ ggml_backend_sycl_device_offload_op,
+    /* .event_new               = */ ggml_backend_sycl_device_event_new,
+    /* .event_free              = */ ggml_backend_sycl_device_event_free,
+    /* .event_synchronize       = */ ggml_backend_sycl_device_event_synchronize,
 };
 
-static ggml_guid_t ggml_backend_sycl_guid() {
-    static ggml_guid guid = { 0x58, 0x05, 0x13, 0x8f, 0xcd, 0x3a, 0x61, 0x9d, 0xe7, 0xcd, 0x98, 0xa9, 0x03, 0xfd, 0x7c, 0x53 };
-    return &guid;
+// backend reg
+
+struct ggml_backend_sycl_reg_context {
+    std::vector devices;
+};
+
+static const char * ggml_backend_sycl_reg_get_name(ggml_backend_reg_t reg) {
+    GGML_UNUSED(reg);
+    return GGML_SYCL_NAME;
+}
+
+static size_t ggml_backend_sycl_reg_get_device_count(ggml_backend_reg_t reg) {
+    ggml_backend_sycl_reg_context * ctx = (ggml_backend_sycl_reg_context *)reg->context;
+    return ctx->devices.size();
+}
+
+static ggml_backend_dev_t ggml_backend_sycl_reg_get_device(ggml_backend_reg_t reg, size_t index) {
+    ggml_backend_sycl_reg_context * ctx = (ggml_backend_sycl_reg_context *)reg->context;
+    GGML_ASSERT(index < ctx->devices.size());
+    return ctx->devices[index];
+}
+
+static void *ggml_backend_sycl_reg_get_proc_address(ggml_backend_reg_t reg, const char *name)
+{
+    GGML_UNUSED(reg);
+    if (strcmp(name, "ggml_backend_split_buffer_type") == 0) {
+        return (void *)ggml_backend_sycl_split_buffer_type;
+    }
+    // SYCL doesn't support registering host memory, left here for reference
+    // "ggml_backend_register_host_buffer"
+    // "ggml_backend_unregister_host_buffer"
+    return nullptr;
+}
+
+static const ggml_backend_reg_i ggml_backend_sycl_reg_interface = {
+    /* .get_name          = */ ggml_backend_sycl_reg_get_name,
+    /* .get_device_count  = */ ggml_backend_sycl_reg_get_device_count,
+    /* .get_device_get    = */ ggml_backend_sycl_reg_get_device,
+    /* .get_proc_address  = */ ggml_backend_sycl_reg_get_proc_address,
+};
+
+
+// backend registry
+
+ggml_backend_reg_t ggml_backend_sycl_reg() {
+    static ggml_backend_reg reg;
+    static bool initialized = false;
+
+    {
+        static std::mutex mutex;
+        std::lock_guard lock(mutex);
+        if (!initialized) {
+            ggml_backend_sycl_reg_context * ctx = new ggml_backend_sycl_reg_context;
+
+            for (int i = 0; i < ggml_sycl_info().device_count; i++) {
+                ggml_backend_sycl_device_context * dev_ctx = new ggml_backend_sycl_device_context;
+                dev_ctx->device = i;
+                dev_ctx->name = GGML_SYCL_NAME + std::to_string(i);
+
+                ggml_sycl_set_device(i);
+
+                dpct::device_info prop;
+                SYCL_CHECK(CHECK_TRY_ERROR(dpct::get_device_info(
+                    prop, dpct::dev_mgr::instance().get_device(i))));
+
+                dev_ctx->description = prop.get_name();
+
+                ggml_backend_dev_t dev = new ggml_backend_device {
+                    /* .interface = */ ggml_backend_sycl_device_interface,
+                    /* .reg       = */ ®,
+                    /* .context   = */ dev_ctx
+                };
+                ctx->devices.push_back(dev);
+            }
+
+            reg = ggml_backend_reg {
+                /* .interface = */ ggml_backend_sycl_reg_interface,
+                /* .context   = */ ctx
+            };
+        }
+
+        initialized = true;
+    }
+
+    return ®
 }
 
 ggml_backend_t ggml_backend_sycl_init(int device) {
@@ -5225,18 +5468,10 @@ ggml_backend_t ggml_backend_sycl_init(int device) {
     ggml_backend_t sycl_backend = new ggml_backend {
         /* .guid      = */ ggml_backend_sycl_guid(),
         /* .interface = */ ggml_backend_sycl_interface,
-        /* .device    = */ nullptr,
+        /* .device    = */ ggml_backend_reg_dev_get(ggml_backend_sycl_reg(), device),
         /* .context   = */ ctx
     };
 
     return sycl_backend;
 }
 
-bool ggml_backend_is_sycl(ggml_backend_t backend) {
-    return backend != NULL && ggml_guid_matches(backend->guid, ggml_backend_sycl_guid());
-}
-
-int ggml_backend_sycl_get_device_count() {
-    GGML_SYCL_DEBUG("[SYCL] call ggml_backend_sycl_get_device_count\n");
-    return ggml_sycl_info().device_count;
-}
diff --git a/src/llama.cpp b/src/llama.cpp
index 0025e94b8..10c975bf4 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -8,9 +8,7 @@
 #include "ggml-alloc.h"
 #include "ggml-backend.h"
 
-#if defined(GGML_USE_SYCL)
-#  include "ggml-sycl.h"
-#elif defined(GGML_USE_KOMPUTE)
+#if defined(GGML_USE_KOMPUTE)
 #   include "ggml-kompute.h"
 #elif defined(GGML_USE_CANN)
 #   include "ggml-cann.h"
@@ -3422,9 +3420,11 @@ struct llama_lora_adapter {
 static int llama_get_device_count(const llama_model & model) {
     int count = (int) model.devices.size();
 
-#if defined(GGML_USE_SYCL)
-    count += ggml_backend_sycl_get_device_count();
-#elif defined(GGML_USE_CANN)
+#if defined(GGML_USE_RPC)
+    count += (int) model.rpc_servers.size();
+#endif
+
+#if defined(GGML_USE_CANN)
     count += ggml_backend_cann_get_device_count();
 #endif
 
@@ -3445,11 +3445,7 @@ static ggml_backend_buffer_type_t llama_default_buffer_type_cpu(const llama_mode
         }
     }
 
-#if defined(GGML_USE_SYCL)
-    if (host_buffer) {
-        buft = ggml_backend_sycl_host_buffer_type();
-    }
-#elif defined(GGML_USE_CANN)
+#if defined(GGML_USE_CANN)
     if (host_buffer) {
         buft = ggml_backend_cann_host_buffer_type();
     }
@@ -3473,9 +3469,7 @@ static ggml_backend_buffer_type_t llama_default_buffer_type_offload(const llama_
     }
     device -= (int)model.devices.size();
 
-#if defined(GGML_USE_SYCL)
-    buft = ggml_backend_sycl_buffer_type(device);
-#elif defined(GGML_USE_KOMPUTE)
+#if defined(GGML_USE_KOMPUTE)
     buft = ggml_backend_kompute_buffer_type(device);
 #elif defined(GGML_USE_CANN)
     buft = ggml_backend_cann_buffer_type(device);
@@ -3505,12 +3499,6 @@ static ggml_backend_buffer_type_t llama_default_buffer_type_split(const llama_mo
         }
     }
 
-#ifdef GGML_USE_SYCL
-    if (ggml_backend_sycl_get_device_count() > 1) {
-        buft = ggml_backend_sycl_split_buffer_type(tensor_split);
-    }
-#endif
-
     if (buft == nullptr) {
         buft = llama_default_buffer_type_offload(model, fallback_gpu);
     }
@@ -3528,12 +3516,7 @@ static size_t llama_get_device_memory(const llama_model & model, int device) {
         return free;
     }
 
-#if defined(GGML_USE_SYCL)
-    size_t total;
-    size_t free;
-    ggml_backend_sycl_get_device_memory(device, &free, &total);
-    return free;
-#elif defined(GGML_USE_CANN)
+#if defined(GGML_USE_CANN)
     size_t total;
     size_t free;
     ggml_backend_cann_get_device_memory(device, &free, &total);
@@ -19096,7 +19079,7 @@ bool llama_supports_mlock(void) {
 }
 
 bool llama_supports_gpu_offload(void) {
-#if defined(GGML_USE_SYCL) || defined(GGML_USE_KOMPUTE)
+#if defined(GGML_USE_KOMPUTE)
     // Defined when llama.cpp is compiled with support for offloading model layers to GPU.
     return true;
 #else
@@ -19428,29 +19411,7 @@ struct llama_context * llama_new_context_with_model(
             main_gpu -= (int)model->devices.size();
         }
 
-#if defined(GGML_USE_SYCL)
-        // with split_mode LLAMA_SPLIT_MODE_NONE or LLAMA_SPLIT_MODE_ROW, only the main GPU backend is used
-        if (model->split_mode == LLAMA_SPLIT_MODE_NONE || model->split_mode == LLAMA_SPLIT_MODE_ROW) {
-            ggml_backend_t backend = ggml_backend_sycl_init(main_gpu);
-            if (backend == nullptr) {
-                LLAMA_LOG_ERROR("%s: failed to initialize SYCL%d backend\n", __func__, main_gpu);
-                llama_free(ctx);
-                return nullptr;
-            }
-            ctx->backends.push_back(backend);
-        } else {
-            // LLAMA_SPLIT_LAYER requires a backend for each GPU
-            for (int i = 0; i < ggml_backend_sycl_get_device_count(); ++i) {
-                ggml_backend_t backend = ggml_backend_sycl_init(i);
-                if (backend == nullptr) {
-                    LLAMA_LOG_ERROR("%s: failed to initialize SYCL%d for No.%d backend\n", __func__, i, i);
-                    llama_free(ctx);
-                    return nullptr;
-                }
-                ctx->backends.push_back(backend);
-            }
-        }
-#elif defined(GGML_USE_KOMPUTE)
+#if defined(GGML_USE_KOMPUTE)
         if (model->n_gpu_layers > 0) {
             auto * backend = ggml_backend_kompute_init(main_gpu);
             if (backend == nullptr) {

From afd9909a6481402844aecefa8a8908afdd7f52f1 Mon Sep 17 00:00:00 2001
From: Radoslav Gerganov 
Date: Fri, 18 Oct 2024 14:33:58 +0300
Subject: [PATCH 36/36] rpc : backend refactoring (#9912)

* rpc : refactor backend

Use structs for RPC request/response messages

* rpc : refactor server
---
 ggml/src/ggml-rpc.cpp | 571 +++++++++++++++++++++++-------------------
 1 file changed, 310 insertions(+), 261 deletions(-)

diff --git a/ggml/src/ggml-rpc.cpp b/ggml/src/ggml-rpc.cpp
index 13c7dd436..f95233284 100644
--- a/ggml/src/ggml-rpc.cpp
+++ b/ggml/src/ggml-rpc.cpp
@@ -58,7 +58,7 @@ struct socket_t {
 };
 
 // ggml_tensor is serialized into rpc_tensor
-#pragma pack(push, 1)
+#pragma pack(1)
 struct rpc_tensor {
     uint64_t id;
     uint32_t type;
@@ -76,7 +76,6 @@ struct rpc_tensor {
 
     char padding[4];
 };
-#pragma pack(pop)
 
 static_assert(sizeof(rpc_tensor) % 8 == 0, "rpc_tensor size must be multiple of 8");
 
@@ -96,6 +95,77 @@ enum rpc_cmd {
     RPC_CMD_COUNT,
 };
 
+#pragma pack(1)
+struct rpc_msg_alloc_buffer_req {
+    uint64_t size;
+};
+
+#pragma pack(1)
+struct rpc_msg_alloc_buffer_rsp {
+    uint64_t remote_ptr;
+    uint64_t remote_size;
+};
+
+#pragma pack(1)
+struct rpc_msg_get_alignment_rsp {
+    uint64_t alignment;
+};
+
+#pragma pack(1)
+struct rpc_msg_get_max_size_rsp {
+    uint64_t max_size;
+};
+
+#pragma pack(1)
+struct rpc_msg_buffer_get_base_req {
+    uint64_t remote_ptr;
+};
+
+#pragma pack(1)
+struct rpc_msg_buffer_get_base_rsp {
+    uint64_t base_ptr;
+};
+
+#pragma pack(1)
+struct rpc_msg_free_buffer_req {
+    uint64_t remote_ptr;
+};
+
+#pragma pack(1)
+struct rpc_msg_buffer_clear_req {
+    uint64_t remote_ptr;
+    uint8_t value;
+};
+
+#pragma pack(1)
+struct rpc_msg_get_tensor_req {
+    rpc_tensor tensor;
+    uint64_t offset;
+    uint64_t size;
+};
+
+#pragma pack(1)
+struct rpc_msg_copy_tensor_req {
+    rpc_tensor src;
+    rpc_tensor dst;
+};
+
+#pragma pack(1)
+struct rpc_msg_copy_tensor_rsp {
+    uint8_t result;
+};
+
+#pragma pack(1)
+struct rpc_msg_graph_compute_rsp {
+    uint8_t result;
+};
+
+#pragma pack(1)
+struct rpc_msg_get_device_memory_rsp {
+    uint64_t free_mem;
+    uint64_t total_mem;
+};
+
 // RPC data structures
 
 static ggml_guid_t ggml_backend_rpc_guid() {
@@ -240,6 +310,38 @@ static bool recv_data(sockfd_t sockfd, void * data, size_t size) {
     return true;
 }
 
+static bool send_msg(sockfd_t sockfd, const void * msg, size_t msg_size) {
+    if (!send_data(sockfd, &msg_size, sizeof(msg_size))) {
+        return false;
+    }
+    return send_data(sockfd, msg, msg_size);
+}
+
+static bool recv_msg(sockfd_t sockfd, void * msg, size_t msg_size) {
+    uint64_t size;
+    if (!recv_data(sockfd, &size, sizeof(size))) {
+        return false;
+    }
+    if (size != msg_size) {
+        return false;
+    }
+    return recv_data(sockfd, msg, msg_size);
+}
+
+static bool recv_msg(sockfd_t sockfd, std::vector & input) {
+    uint64_t size;
+    if (!recv_data(sockfd, &size, sizeof(size))) {
+        return false;
+    }
+    try {
+        input.resize(size);
+    } catch (const std::bad_alloc & e) {
+        fprintf(stderr, "Failed to allocate input buffer of size %" PRIu64 "\n", size);
+        return false;
+    }
+    return recv_data(sockfd, input.data(), size);
+}
+
 static bool parse_endpoint(const std::string & endpoint, std::string & host, int & port) {
     size_t pos = endpoint.find(':');
     if (pos == std::string::npos) {
@@ -252,28 +354,27 @@ static bool parse_endpoint(const std::string & endpoint, std::string & host, int
 
 // RPC request : | rpc_cmd (1 byte) | request_size (8 bytes) | request_data (request_size bytes) |
 // RPC response: | response_size (8 bytes) | response_data (response_size bytes) |
-static bool send_rpc_cmd(const std::shared_ptr & sock, enum rpc_cmd cmd, const std::vector & input, std::vector & output) {
+static bool send_rpc_cmd(const std::shared_ptr & sock, enum rpc_cmd cmd, const void * input, size_t input_size, void * output, size_t output_size) {
     uint8_t cmd_byte = cmd;
     if (!send_data(sock->fd, &cmd_byte, sizeof(cmd_byte))) {
         return false;
     }
-    uint64_t input_size = input.size();
     if (!send_data(sock->fd, &input_size, sizeof(input_size))) {
         return false;
     }
-    if (!send_data(sock->fd, input.data(), input.size())) {
+    if (!send_data(sock->fd, input, input_size)) {
         return false;
     }
-    uint64_t output_size;
-    if (!recv_data(sock->fd, &output_size, sizeof(output_size))) {
+    // TODO: currently the output_size is always known, do we need support for commands with variable output size?
+    // even if we do, we can skip sending output_size from the server for commands with known output size
+    uint64_t out_size;
+    if (!recv_data(sock->fd, &out_size, sizeof(out_size))) {
         return false;
     }
-    if (output_size == 0) {
-        output.clear();
-        return true;
+    if (out_size != output_size) {
+        return false;
     }
-    output.resize(output_size);
-    if (!recv_data(sock->fd, output.data(), output_size)) {
+    if (!recv_data(sock->fd, output, output_size)) {
         return false;
     }
     return true;
@@ -326,14 +427,9 @@ static const char * ggml_backend_rpc_buffer_get_name(ggml_backend_buffer_t buffe
 
 static void ggml_backend_rpc_buffer_free_buffer(ggml_backend_buffer_t buffer) {
     ggml_backend_rpc_buffer_context * ctx = (ggml_backend_rpc_buffer_context *)buffer->context;
-    // input serialization format: | remote_ptr (8 bytes) |
-    std::vector input(sizeof(uint64_t), 0);
-    uint64_t remote_ptr = ctx->remote_ptr;
-    memcpy(input.data(), &remote_ptr, sizeof(remote_ptr));
-    std::vector output;
-    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_FREE_BUFFER, input, output);
+    rpc_msg_free_buffer_req request = {ctx->remote_ptr};
+    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_FREE_BUFFER, &request, sizeof(request), nullptr, 0);
     GGML_ASSERT(status);
-    GGML_ASSERT(output.empty());
     delete ctx;
 }
 
@@ -342,20 +438,13 @@ static void * ggml_backend_rpc_buffer_get_base(ggml_backend_buffer_t buffer) {
     if (ctx->base_cache.find(buffer) != ctx->base_cache.end()) {
         return ctx->base_cache[buffer];
     }
-    // input serialization format: | remote_ptr (8 bytes) |
-    std::vector input(sizeof(uint64_t), 0);
-    uint64_t remote_ptr = ctx->remote_ptr;
-    memcpy(input.data(), &remote_ptr, sizeof(remote_ptr));
-    std::vector output;
-    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_BUFFER_GET_BASE, input, output);
+    rpc_msg_buffer_get_base_req request = {ctx->remote_ptr};
+    rpc_msg_buffer_get_base_rsp response;
+    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_BUFFER_GET_BASE, &request, sizeof(request), &response, sizeof(response));
     GGML_ASSERT(status);
-    GGML_ASSERT(output.size() == sizeof(uint64_t));
-    // output serialization format: | base_ptr (8 bytes) |
-    uint64_t base_ptr;
-    memcpy(&base_ptr, output.data(), sizeof(base_ptr));
-    void * base = reinterpret_cast(base_ptr);
-    ctx->base_cache[buffer] = base;
-    return base;
+    void * base_ptr = reinterpret_cast(response.base_ptr);
+    ctx->base_cache[buffer] = base_ptr;
+    return base_ptr;
 }
 
 static rpc_tensor serialize_tensor(const ggml_tensor * tensor) {
@@ -405,26 +494,18 @@ static void ggml_backend_rpc_buffer_set_tensor(ggml_backend_buffer_t buffer, ggm
     memcpy(input.data(), &rpc_tensor, sizeof(rpc_tensor));
     memcpy(input.data() + sizeof(rpc_tensor), &offset, sizeof(offset));
     memcpy(input.data() + sizeof(rpc_tensor) + sizeof(offset), data, size);
-    std::vector output;
-    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_SET_TENSOR, input, output);
+    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_SET_TENSOR, input.data(), input.size(), nullptr, 0);
     GGML_ASSERT(status);
 }
 
 static void ggml_backend_rpc_buffer_get_tensor(ggml_backend_buffer_t buffer, const ggml_tensor * tensor, void * data, size_t offset, size_t size) {
     ggml_backend_rpc_buffer_context * ctx = (ggml_backend_rpc_buffer_context *)buffer->context;
-    // input serialization format: | rpc_tensor | offset (8 bytes) | size (8 bytes) |
-    int input_size = sizeof(rpc_tensor) + 2*sizeof(uint64_t);
-    std::vector input(input_size, 0);
-    rpc_tensor rpc_tensor = serialize_tensor(tensor);
-    memcpy(input.data(), &rpc_tensor, sizeof(rpc_tensor));
-    memcpy(input.data() + sizeof(rpc_tensor), &offset, sizeof(offset));
-    memcpy(input.data() + sizeof(rpc_tensor) + sizeof(offset), &size, sizeof(size));
-    std::vector output;
-    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_GET_TENSOR, input, output);
+    rpc_msg_get_tensor_req request;
+    request.tensor = serialize_tensor(tensor);
+    request.offset = offset;
+    request.size = size;
+    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_GET_TENSOR, &request, sizeof(request), data, size);
     GGML_ASSERT(status);
-    GGML_ASSERT(output.size() == size);
-    // output serialization format: | data (size bytes) |
-    memcpy(data, output.data(), size);
 }
 
 static bool ggml_backend_rpc_buffer_cpy_tensor(ggml_backend_buffer_t buffer, const ggml_tensor * src, ggml_tensor * dst) {
@@ -437,30 +518,19 @@ static bool ggml_backend_rpc_buffer_cpy_tensor(ggml_backend_buffer_t buffer, con
         return false;
     }
     ggml_backend_rpc_buffer_context * ctx = (ggml_backend_rpc_buffer_context *)buffer->context;
-    // input serialization format: | rpc_tensor src | rpc_tensor dst |
-    int input_size = 2*sizeof(rpc_tensor);
-    std::vector input(input_size, 0);
-    rpc_tensor rpc_src = serialize_tensor(src);
-    rpc_tensor rpc_dst = serialize_tensor(dst);
-    memcpy(input.data(), &rpc_src, sizeof(rpc_src));
-    memcpy(input.data() + sizeof(rpc_src), &rpc_dst, sizeof(rpc_dst));
-    std::vector output;
-    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_COPY_TENSOR, input, output);
+    rpc_msg_copy_tensor_req request;
+    request.src = serialize_tensor(src);
+    request.dst = serialize_tensor(dst);
+    rpc_msg_copy_tensor_rsp response;
+    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_COPY_TENSOR, &request, sizeof(request), &response, sizeof(response));
     GGML_ASSERT(status);
-    // output serialization format: | result (1 byte) |
-    GGML_ASSERT(output.size() == 1);
-    return output[0];
+    return response.result;
 }
 
 static void ggml_backend_rpc_buffer_clear(ggml_backend_buffer_t buffer, uint8_t value) {
     ggml_backend_rpc_buffer_context * ctx = (ggml_backend_rpc_buffer_context *)buffer->context;
-    // serialization format: | bufptr (8 bytes) | value (1 byte) |
-    int input_size = sizeof(uint64_t) + sizeof(uint8_t);
-    std::vector input(input_size, 0);
-    memcpy(input.data(), &ctx->remote_ptr, sizeof(ctx->remote_ptr));
-    memcpy(input.data() + sizeof(ctx->remote_ptr), &value, sizeof(value));
-    std::vector output;
-    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_BUFFER_CLEAR, input, output);
+    rpc_msg_buffer_clear_req request = {ctx->remote_ptr, value};
+    bool status = send_rpc_cmd(ctx->sock, RPC_CMD_BUFFER_CLEAR, &request, sizeof(request), nullptr, 0);
     GGML_ASSERT(status);
 }
 
@@ -484,25 +554,16 @@ static const char * ggml_backend_rpc_buffer_type_name(ggml_backend_buffer_type_t
 
 static ggml_backend_buffer_t ggml_backend_rpc_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
     ggml_backend_rpc_buffer_type_context * buft_ctx = (ggml_backend_rpc_buffer_type_context *)buft->context;
-    // input serialization format: | size (8 bytes) |
-    int input_size = sizeof(uint64_t);
-    std::vector input(input_size, 0);
-    memcpy(input.data(), &size, sizeof(size));
-    std::vector output;
+    rpc_msg_alloc_buffer_req request = {size};
+    rpc_msg_alloc_buffer_rsp response;
     auto sock = get_socket(buft_ctx->endpoint);
-    bool status = send_rpc_cmd(sock, RPC_CMD_ALLOC_BUFFER, input, output);
+    bool status = send_rpc_cmd(sock, RPC_CMD_ALLOC_BUFFER, &request, sizeof(request), &response, sizeof(response));
     GGML_ASSERT(status);
-    GGML_ASSERT(output.size() == 2*sizeof(uint64_t));
-    // output serialization format: | remote_ptr (8 bytes) | remote_size (8 bytes) |
-    uint64_t remote_ptr;
-    memcpy(&remote_ptr, output.data(), sizeof(remote_ptr));
-    size_t remote_size;
-    memcpy(&remote_size, output.data() + sizeof(uint64_t), sizeof(remote_size));
-    if (remote_ptr != 0) {
+    if (response.remote_ptr != 0) {
         ggml_backend_buffer_t buffer = ggml_backend_buffer_init(buft,
             ggml_backend_rpc_buffer_interface,
-            new ggml_backend_rpc_buffer_context{sock, {}, remote_ptr, "RPC[" + std::string(buft_ctx->endpoint) + "]"},
-            remote_size);
+            new ggml_backend_rpc_buffer_context{sock, {}, response.remote_ptr, "RPC[" + std::string(buft_ctx->endpoint) + "]"},
+            response.remote_size);
         return buffer;
     } else {
         return nullptr;
@@ -510,16 +571,10 @@ static ggml_backend_buffer_t ggml_backend_rpc_buffer_type_alloc_buffer(ggml_back
 }
 
 static size_t get_alignment(const std::shared_ptr & sock) {
-    // input serialization format: | 0 bytes |
-    std::vector input;
-    std::vector output;
-    bool status = send_rpc_cmd(sock, RPC_CMD_GET_ALIGNMENT, input, output);
+    rpc_msg_get_alignment_rsp response;
+    bool status = send_rpc_cmd(sock, RPC_CMD_GET_ALIGNMENT, nullptr, 0, &response, sizeof(response));
     GGML_ASSERT(status);
-    GGML_ASSERT(output.size() == sizeof(uint64_t));
-    // output serialization format: | alignment (8 bytes) |
-    uint64_t alignment;
-    memcpy(&alignment, output.data(), sizeof(alignment));
-    return alignment;
+    return response.alignment;
 }
 
 static size_t ggml_backend_rpc_buffer_type_get_alignment(ggml_backend_buffer_type_t buft) {
@@ -528,16 +583,10 @@ static size_t ggml_backend_rpc_buffer_type_get_alignment(ggml_backend_buffer_typ
 }
 
 static size_t get_max_size(const std::shared_ptr & sock) {
-    // input serialization format: | 0 bytes |
-    std::vector input;
-    std::vector output;
-    bool status = send_rpc_cmd(sock, RPC_CMD_GET_MAX_SIZE, input, output);
+    rpc_msg_get_max_size_rsp response;
+    bool status = send_rpc_cmd(sock, RPC_CMD_GET_MAX_SIZE, nullptr, 0, &response, sizeof(response));
     GGML_ASSERT(status);
-    GGML_ASSERT(output.size() == sizeof(uint64_t));
-    // output serialization format: | max_size (8 bytes) |
-    uint64_t max_size;
-    memcpy(&max_size, output.data(), sizeof(max_size));
-    return max_size;
+    return response.max_size;
 }
 
 static size_t ggml_backend_rpc_get_max_size(ggml_backend_buffer_type_t buft) {
@@ -622,12 +671,11 @@ static enum ggml_status ggml_backend_rpc_graph_compute(ggml_backend_t backend, g
     ggml_backend_rpc_context * rpc_ctx = (ggml_backend_rpc_context *)backend->context;
     std::vector input;
     serialize_graph(cgraph, input);
-    std::vector output;
+    rpc_msg_graph_compute_rsp response;
     auto sock = get_socket(rpc_ctx->endpoint);
-    bool status = send_rpc_cmd(sock, RPC_CMD_GRAPH_COMPUTE, input, output);
+    bool status = send_rpc_cmd(sock, RPC_CMD_GRAPH_COMPUTE, input.data(), input.size(), &response, sizeof(response));
     GGML_ASSERT(status);
-    GGML_ASSERT(output.size() == 1);
-    return (enum ggml_status)output[0];
+    return (enum ggml_status)response.result;
 }
 
 static ggml_backend_i ggml_backend_rpc_interface = {
@@ -702,19 +750,11 @@ GGML_API bool ggml_backend_is_rpc(ggml_backend_t backend) {
 }
 
 static void get_device_memory(const std::shared_ptr & sock, size_t * free, size_t * total) {
-    // input serialization format: | 0 bytes |
-    std::vector input;
-    std::vector output;
-    bool status = send_rpc_cmd(sock, RPC_CMD_GET_DEVICE_MEMORY, input, output);
+    rpc_msg_get_device_memory_rsp response;
+    bool status = send_rpc_cmd(sock, RPC_CMD_GET_DEVICE_MEMORY, nullptr, 0, &response, sizeof(response));
     GGML_ASSERT(status);
-    GGML_ASSERT(output.size() == 2*sizeof(uint64_t));
-    // output serialization format: | free (8 bytes) | total (8 bytes) |
-    uint64_t free_mem;
-    memcpy(&free_mem, output.data(), sizeof(free_mem));
-    uint64_t total_mem;
-    memcpy(&total_mem, output.data() + sizeof(uint64_t), sizeof(total_mem));
-    *free = free_mem;
-    *total = total_mem;
+    *free = response.free_mem;
+    *total = response.total_mem;
 }
 
 GGML_API void ggml_backend_rpc_get_device_memory(const char * endpoint, size_t * free, size_t * total) {
@@ -734,16 +774,16 @@ public:
     rpc_server(ggml_backend_t backend) : backend(backend) {}
     ~rpc_server();
 
-    bool alloc_buffer(const std::vector & input, std::vector & output);
-    void get_alignment(std::vector & output);
-    void get_max_size(std::vector & output);
-    bool buffer_get_base(const std::vector & input, std::vector & output);
-    bool free_buffer(const std::vector & input);
-    bool buffer_clear(const std::vector & input);
+    void alloc_buffer(const rpc_msg_alloc_buffer_req & request, rpc_msg_alloc_buffer_rsp & response);
+    void get_alignment(rpc_msg_get_alignment_rsp & response);
+    void get_max_size(rpc_msg_get_max_size_rsp & response);
+    bool buffer_get_base(const rpc_msg_buffer_get_base_req & request, rpc_msg_buffer_get_base_rsp & response);
+    bool free_buffer(const rpc_msg_free_buffer_req & request);
+    bool buffer_clear(const rpc_msg_buffer_clear_req & request);
     bool set_tensor(const std::vector & input);
-    bool get_tensor(const std::vector & input, std::vector & output);
-    bool copy_tensor(const std::vector & input, std::vector & output);
-    bool graph_compute(const std::vector & input, std::vector & output);
+    bool get_tensor(const rpc_msg_get_tensor_req & request, std::vector & response);
+    bool copy_tensor(const rpc_msg_copy_tensor_req & request, rpc_msg_copy_tensor_rsp & response);
+    bool graph_compute(const std::vector & input, rpc_msg_graph_compute_rsp & response);
 
 private:
     ggml_tensor * deserialize_tensor(struct ggml_context * ctx, const rpc_tensor * tensor);
@@ -757,80 +797,50 @@ private:
     std::unordered_set buffers;
 };
 
-bool rpc_server::alloc_buffer(const std::vector & input, std::vector & output) {
-    // input serialization format: | size (8 bytes) |
-    if (input.size() != sizeof(uint64_t)) {
-        return false;
-    }
-    uint64_t size;
-    memcpy(&size, input.data(), sizeof(size));
+void rpc_server::alloc_buffer(const rpc_msg_alloc_buffer_req & request, rpc_msg_alloc_buffer_rsp & response) {
     ggml_backend_buffer_type_t buft = ggml_backend_get_default_buffer_type(backend);
-    ggml_backend_buffer_t buffer = ggml_backend_buft_alloc_buffer(buft, size);
-    uint64_t remote_ptr = 0;
-    uint64_t remote_size = 0;
+    ggml_backend_buffer_t buffer = ggml_backend_buft_alloc_buffer(buft, request.size);
+    response.remote_ptr = 0;
+    response.remote_size = 0;
     if (buffer != nullptr) {
-        remote_ptr = reinterpret_cast(buffer);
-        remote_size = buffer->size;
-        GGML_PRINT_DEBUG("[%s] size: %" PRIu64 " -> remote_ptr: %" PRIx64 ", remote_size: %" PRIu64 "\n", __func__, size, remote_ptr, remote_size);
+        response.remote_ptr = reinterpret_cast(buffer);
+        response.remote_size = buffer->size;
+        GGML_PRINT_DEBUG("[%s] size: %" PRIu64 " -> remote_ptr: %" PRIx64 ", remote_size: %" PRIu64 "\n", __func__, request.size, response.remote_ptr, response.remote_size);
         buffers.insert(buffer);
     } else {
-        GGML_PRINT_DEBUG("[%s] size: %" PRIu64 " -> failed\n", __func__, size);
+        GGML_PRINT_DEBUG("[%s] size: %" PRIu64 " -> failed\n", __func__, request.size);
     }
-    // output serialization format: | remote_ptr (8 bytes) | remote_size (8 bytes) |
-    output.resize(2*sizeof(uint64_t), 0);
-    memcpy(output.data(), &remote_ptr, sizeof(remote_ptr));
-    memcpy(output.data() + sizeof(uint64_t), &remote_size, sizeof(remote_size));
-    return true;
 }
 
-void rpc_server::get_alignment(std::vector & output) {
+void rpc_server::get_alignment(rpc_msg_get_alignment_rsp & response) {
     ggml_backend_buffer_type_t buft = ggml_backend_get_default_buffer_type(backend);
     size_t alignment = ggml_backend_buft_get_alignment(buft);
     GGML_PRINT_DEBUG("[%s] alignment: %lu\n", __func__, alignment);
-    // output serialization format: | alignment (8 bytes) |
-    output.resize(sizeof(uint64_t), 0);
-    memcpy(output.data(), &alignment, sizeof(alignment));
+    response.alignment = alignment;
 }
 
-void rpc_server::get_max_size(std::vector & output) {
+void rpc_server::get_max_size(rpc_msg_get_max_size_rsp & response) {
     ggml_backend_buffer_type_t buft = ggml_backend_get_default_buffer_type(backend);
     size_t max_size = ggml_backend_buft_get_max_size(buft);
     GGML_PRINT_DEBUG("[%s] max_size: %lu\n", __func__, max_size);
-    // output serialization format: | max_size (8 bytes) |
-    output.resize(sizeof(uint64_t), 0);
-    memcpy(output.data(), &max_size, sizeof(max_size));
+    response.max_size = max_size;
 }
 
-bool rpc_server::buffer_get_base(const std::vector & input, std::vector & output) {
-    // input serialization format: | remote_ptr (8 bytes) |
-    if (input.size() != sizeof(uint64_t)) {
-        return false;
-    }
-    uint64_t remote_ptr;
-    memcpy(&remote_ptr, input.data(), sizeof(remote_ptr));
-    GGML_PRINT_DEBUG("[%s] remote_ptr: %" PRIx64 "\n", __func__, remote_ptr);
-    ggml_backend_buffer_t buffer = reinterpret_cast(remote_ptr);
+bool rpc_server::buffer_get_base(const rpc_msg_buffer_get_base_req & request, rpc_msg_buffer_get_base_rsp & response) {
+    GGML_PRINT_DEBUG("[%s] remote_ptr: %" PRIx64 "\n", __func__, request.remote_ptr);
+    ggml_backend_buffer_t buffer = reinterpret_cast(request.remote_ptr);
     if (buffers.find(buffer) == buffers.end()) {
         GGML_PRINT_DEBUG("[%s] buffer not found\n", __func__);
         return false;
     }
     void * base = ggml_backend_buffer_get_base(buffer);
-    // output serialization format: | base_ptr (8 bytes) |
-    uint64_t base_ptr = reinterpret_cast(base);
-    output.resize(sizeof(uint64_t), 0);
-    memcpy(output.data(), &base_ptr, sizeof(base_ptr));
+    response.base_ptr = reinterpret_cast(base);
     return true;
 }
 
-bool rpc_server::free_buffer(const std::vector & input) {
-    // input serialization format: | remote_ptr (8 bytes) |
-    if (input.size() != sizeof(uint64_t)) {
-        return false;
-    }
-    uint64_t remote_ptr;
-    memcpy(&remote_ptr, input.data(), sizeof(remote_ptr));
-    GGML_PRINT_DEBUG("[%s] remote_ptr: %" PRIx64 "\n", __func__, remote_ptr);
-    ggml_backend_buffer_t buffer = reinterpret_cast(remote_ptr);
+bool rpc_server::free_buffer(const rpc_msg_free_buffer_req & request) {
+    GGML_PRINT_DEBUG("[%s] remote_ptr: %" PRIx64 "\n", __func__, request.remote_ptr);
+    ggml_backend_buffer_t buffer = reinterpret_cast(request.remote_ptr);
     if (buffers.find(buffer) == buffers.end()) {
         GGML_PRINT_DEBUG("[%s] buffer not found\n", __func__);
         return false;
@@ -840,22 +850,14 @@ bool rpc_server::free_buffer(const std::vector & input) {
     return true;
 }
 
-bool rpc_server::buffer_clear(const std::vector & input) {
-    // input serialization format: | remote_ptr (8 bytes) | value (1 byte) |
-    if (input.size() != sizeof(uint64_t) + sizeof(uint8_t)) {
-        return false;
-    }
-    uint64_t remote_ptr;
-    memcpy(&remote_ptr, input.data(), sizeof(remote_ptr));
-    uint8_t value;
-    memcpy(&value, input.data() + sizeof(uint64_t), sizeof(value));
-    GGML_PRINT_DEBUG("[%s] remote_ptr: %" PRIx64 ", value: %u\n", __func__, remote_ptr, value);
-    ggml_backend_buffer_t buffer = reinterpret_cast(remote_ptr);
+bool rpc_server::buffer_clear(const rpc_msg_buffer_clear_req & request) {
+    GGML_PRINT_DEBUG("[%s] remote_ptr: %" PRIx64 ", value: %u\n", __func__, request.remote_ptr, request.value);
+    ggml_backend_buffer_t buffer = reinterpret_cast(request.remote_ptr);
     if (buffers.find(buffer) == buffers.end()) {
         GGML_PRINT_DEBUG("[%s] buffer not found\n", __func__);
         return false;
     }
-    ggml_backend_buffer_clear(buffer, value);
+    ggml_backend_buffer_clear(buffer, request.value);
     return true;
 }
 
@@ -930,74 +932,55 @@ bool rpc_server::set_tensor(const std::vector & input) {
     return true;
 }
 
-bool rpc_server::get_tensor(const std::vector & input, std::vector & output) {
-    // serialization format: | rpc_tensor | offset (8 bytes) | size (8 bytes) |
-    if (input.size() != sizeof(rpc_tensor) + 2*sizeof(uint64_t)) {
-        return false;
-    }
-    const rpc_tensor * in_tensor = (const rpc_tensor *)input.data();
-    uint64_t offset;
-    memcpy(&offset, input.data() + sizeof(rpc_tensor), sizeof(offset));
-    uint64_t size;
-    memcpy(&size, input.data() + sizeof(rpc_tensor) + sizeof(offset), sizeof(size));
-
+bool rpc_server::get_tensor(const rpc_msg_get_tensor_req & request, std::vector & response) {
     struct ggml_init_params params {
         /*.mem_size   =*/ ggml_tensor_overhead(),
         /*.mem_buffer =*/ NULL,
         /*.no_alloc   =*/ true,
     };
     struct ggml_context * ctx = ggml_init(params);
-    ggml_tensor * tensor = deserialize_tensor(ctx, in_tensor);
+    ggml_tensor * tensor = deserialize_tensor(ctx, &request.tensor);
     if (tensor == nullptr) {
         GGML_PRINT_DEBUG("[%s] error deserializing tensor\n", __func__);
         ggml_free(ctx);
         return false;
     }
-    GGML_PRINT_DEBUG("[%s] buffer: %p, data: %p, offset: %" PRIu64 ", size: %" PRIu64 "\n", __func__, (void*)tensor->buffer, tensor->data, offset, size);
+    GGML_PRINT_DEBUG("[%s] buffer: %p, data: %p, offset: %" PRIu64 ", size: %" PRIu64 "\n", __func__, (void*)tensor->buffer, tensor->data, request.offset, request.size);
 
     // sanitize tensor->data
     {
         const size_t p0 = (size_t) ggml_backend_buffer_get_base(tensor->buffer);
         const size_t p1 = p0 + ggml_backend_buffer_get_size(tensor->buffer);
 
-        if (in_tensor->data + offset < p0 || in_tensor->data + offset >= p1 || size > (p1 - in_tensor->data - offset)) {
-            GGML_ABORT("[%s] tensor->data out of bounds\n", __func__);
+        if (request.tensor.data + request.offset < p0 ||
+            request.tensor.data + request.offset >= p1 ||
+            request.size > (p1 - request.tensor.data - request.offset)) {
+                GGML_ABORT("[%s] tensor->data out of bounds\n", __func__);
         }
     }
 
-    // output serialization format: | data (size bytes) |
-    output.resize(size, 0);
-    ggml_backend_tensor_get(tensor, output.data(), offset, size);
+    response.resize(request.size, 0);
+    ggml_backend_tensor_get(tensor, response.data(), request.offset, request.size);
     ggml_free(ctx);
     return true;
 }
 
-bool rpc_server::copy_tensor(const std::vector & input, std::vector & output) {
-    // serialization format: | rpc_tensor src | rpc_tensor dst |
-    if (input.size() != 2*sizeof(rpc_tensor)) {
-        return false;
-    }
-    const rpc_tensor * rpc_src = (const rpc_tensor *)input.data();
-    const rpc_tensor * rpc_dst = (const rpc_tensor *)(input.data() + sizeof(rpc_src));
-
+bool rpc_server::copy_tensor(const rpc_msg_copy_tensor_req & request, rpc_msg_copy_tensor_rsp & response) {
     struct ggml_init_params params {
         /*.mem_size   =*/ 2*ggml_tensor_overhead(),
         /*.mem_buffer =*/ NULL,
         /*.no_alloc   =*/ true,
     };
     struct ggml_context * ctx = ggml_init(params);
-    ggml_tensor * src = deserialize_tensor(ctx, rpc_src);
-    ggml_tensor * dst = deserialize_tensor(ctx, rpc_dst);
+    ggml_tensor * src = deserialize_tensor(ctx, &request.src);
+    ggml_tensor * dst = deserialize_tensor(ctx, &request.dst);
     if (src == nullptr || dst == nullptr) {
         GGML_PRINT_DEBUG("[%s] error deserializing tensors\n", __func__);
         ggml_free(ctx);
         return false;
     }
     GGML_PRINT_DEBUG("[%s] src->buffer: %p, dst->buffer: %p\n", __func__, (void*)src->buffer, (void*)dst->buffer);
-    bool result = ggml_backend_buffer_copy_tensor(src, dst);
-    // output serialization format: | result (1 byte) |
-    output.resize(1, 0);
-    output[0] = result;
+    response.result = ggml_backend_buffer_copy_tensor(src, dst);
     ggml_free(ctx);
     return true;
 }
@@ -1026,7 +1009,7 @@ ggml_tensor * rpc_server::create_node(uint64_t id,
     return result;
 }
 
-bool rpc_server::graph_compute(const std::vector & input, std::vector & output) {
+bool rpc_server::graph_compute(const std::vector & input, rpc_msg_graph_compute_rsp & response) {
     // serialization format:
     // | n_nodes (4 bytes) | nodes (n_nodes * sizeof(uint64_t) | n_tensors (4 bytes) | tensors (n_tensors * sizeof(rpc_tensor)) |
     if (input.size() < sizeof(uint32_t)) {
@@ -1066,9 +1049,7 @@ bool rpc_server::graph_compute(const std::vector & input, std::vectornodes[i] = create_node(id, ctx, tensor_ptrs, tensor_map);
     }
     ggml_status status = ggml_backend_graph_compute(backend, graph);
-    // output serialization format: | status (1 byte) |
-    output.resize(1, 0);
-    output[0] = status;
+    response.result = status;
     ggml_free(ctx);
     return true;
 }
@@ -1091,85 +1072,153 @@ static void rpc_serve_client(ggml_backend_t backend, sockfd_t sockfd, size_t fre
             fprintf(stderr, "Unknown command: %d\n", cmd);
             break;
         }
-        std::vector input;
-        std::vector output;
-        uint64_t input_size;
-        if (!recv_data(sockfd, &input_size, sizeof(input_size))) {
-            break;
-        }
-        try {
-            input.resize(input_size);
-        } catch (const std::bad_alloc & e) {
-            fprintf(stderr, "Failed to allocate input buffer of size %" PRIu64 "\n", input_size);
-            break;
-        }
-        if (!recv_data(sockfd, input.data(), input_size)) {
-            break;
-        }
-        bool ok = true;
         switch (cmd) {
             case RPC_CMD_ALLOC_BUFFER: {
-                ok = server.alloc_buffer(input, output);
+                rpc_msg_alloc_buffer_req request;
+                if (!recv_msg(sockfd, &request, sizeof(request))) {
+                    return;
+                }
+                rpc_msg_alloc_buffer_rsp response;
+                server.alloc_buffer(request, response);
+                if (!send_msg(sockfd, &response, sizeof(response))) {
+                    return;
+                }
                 break;
             }
             case RPC_CMD_GET_ALIGNMENT: {
-                server.get_alignment(output);
+                if (!recv_msg(sockfd, nullptr, 0)) {
+                    return;
+                }
+                rpc_msg_get_alignment_rsp response;
+                server.get_alignment(response);
+                if (!send_msg(sockfd, &response, sizeof(response))) {
+                    return;
+                }
                 break;
             }
             case RPC_CMD_GET_MAX_SIZE: {
-                server.get_max_size(output);
+                if (!recv_msg(sockfd, nullptr, 0)) {
+                    return;
+                }
+                rpc_msg_get_max_size_rsp response;
+                server.get_max_size(response);
+                if (!send_msg(sockfd, &response, sizeof(response))) {
+                    return;
+                }
                 break;
             }
             case RPC_CMD_BUFFER_GET_BASE: {
-                ok = server.buffer_get_base(input, output);
+                rpc_msg_buffer_get_base_req request;
+                if (!recv_msg(sockfd, &request, sizeof(request))) {
+                    return;
+                }
+                rpc_msg_buffer_get_base_rsp response;
+                if (!server.buffer_get_base(request, response)) {
+                    return;
+                }
+                if (!send_msg(sockfd, &response, sizeof(response))) {
+                    return;
+                }
                 break;
             }
             case RPC_CMD_FREE_BUFFER: {
-                ok = server.free_buffer(input);
+                rpc_msg_free_buffer_req request;
+                if (!recv_msg(sockfd, &request, sizeof(request))) {
+                    return;
+                }
+                if (!server.free_buffer(request)) {
+                    return;
+                }
+                if (!send_msg(sockfd, nullptr, 0)) {
+                    return;
+                }
                 break;
             }
             case RPC_CMD_BUFFER_CLEAR: {
-                ok = server.buffer_clear(input);
+                rpc_msg_buffer_clear_req request;
+                if (!recv_msg(sockfd, &request, sizeof(request))) {
+                    return;
+                }
+                if (!server.buffer_clear(request)) {
+                    return;
+                }
+                if (!send_msg(sockfd, nullptr, 0)) {
+                    return;
+                }
                 break;
             }
             case RPC_CMD_SET_TENSOR: {
-                ok = server.set_tensor(input);
+                std::vector input;
+                if (!recv_msg(sockfd, input)) {
+                    return;
+                }
+                if (!server.set_tensor(input)) {
+                    return;
+                }
+                if (!send_msg(sockfd, nullptr, 0)) {
+                    return;
+                }
                 break;
             }
             case RPC_CMD_GET_TENSOR: {
-                ok = server.get_tensor(input, output);
+                rpc_msg_get_tensor_req request;
+                if (!recv_msg(sockfd, &request, sizeof(request))) {
+                    return;
+                }
+                std::vector response;
+                if (!server.get_tensor(request, response)) {
+                    return;
+                }
+                if (!send_msg(sockfd, response.data(), response.size())) {
+                    return;
+                }
                 break;
             }
             case RPC_CMD_COPY_TENSOR: {
-                ok = server.copy_tensor(input, output);
+                rpc_msg_copy_tensor_req request;
+                if (!recv_msg(sockfd, &request, sizeof(request))) {
+                    return;
+                }
+                rpc_msg_copy_tensor_rsp response;
+                if (!server.copy_tensor(request, response)) {
+                    return;
+                }
+                if (!send_msg(sockfd, &response, sizeof(response))) {
+                    return;
+                }
                 break;
             }
             case RPC_CMD_GRAPH_COMPUTE: {
-                ok = server.graph_compute(input, output);
+                std::vector input;
+                if (!recv_msg(sockfd, input)) {
+                    return;
+                }
+                rpc_msg_graph_compute_rsp response;
+                if (!server.graph_compute(input, response)) {
+                    return;
+                }
+                if (!send_msg(sockfd, &response, sizeof(response))) {
+                    return;
+                }
                 break;
             }
             case RPC_CMD_GET_DEVICE_MEMORY: {
-                // output serialization format: | free (8 bytes) | total (8 bytes) |
-                output.resize(2*sizeof(uint64_t), 0);
-                memcpy(output.data(), &free_mem, sizeof(free_mem));
-                memcpy(output.data() + sizeof(uint64_t), &total_mem, sizeof(total_mem));
+                if (!recv_msg(sockfd, nullptr, 0)) {
+                    return;
+                }
+                rpc_msg_get_device_memory_rsp response;
+                response.free_mem = free_mem;
+                response.total_mem = total_mem;
+                if (!send_msg(sockfd, &response, sizeof(response))) {
+                    return;
+                }
                 break;
             }
             default: {
                 fprintf(stderr, "Unknown command: %d\n", cmd);
-                ok = false;
+                return;
             }
         }
-        if (!ok) {
-            break;
-        }
-        uint64_t output_size = output.size();
-        if (!send_data(sockfd, &output_size, sizeof(output_size))) {
-            break;
-        }
-        if (!send_data(sockfd, output.data(), output_size)) {
-            break;
-        }
     }
 }