vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge remote-tracking branch 'upstream/master'

Browse source 
# Conflicts:
#	examples/server/README.md
This commit is contained in:
Randall Fitzgerald 2023-06-15 17:12:29 -04:00
commit 488c62acf9
11 changed files with 547 additions and 457 deletions
Download patch file Download diff file Expand all files Collapse all files

1
.gitignore vendored
View file

@ -32,6 +32,7 @@ models/*
/result
/perplexity
/embedding
/train-text-from-scratch
/benchmark-matmult
/vdot
/server

17
CMakeLists.txt
View file

@ -163,12 +163,23 @@ if (LLAMA_BLAS)
if (BLAS_FOUND)
message(STATUS "BLAS found, Libraries: ${BLAS_LIBRARIES}")
# BLAS_INCLUDE_DIRS is missing in FindBLAS.cmake.
# see https://gitlab.kitware.com/cmake/cmake/-/issues/20268
find_path(BLAS_INCLUDE_DIRS
NAMES cblas.h
HINTS
/usr/include
/usr/local/include
/usr/include/openblas
)
message(STATUS "BLAS found, Includes: ${BLAS_INCLUDE_DIRS}")
add_compile_options(${BLAS_LINKER_FLAGS})
add_compile_definitions(GGML_USE_OPENBLAS)
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} ${BLAS_LIBRARIES})
set(LLAMA_EXTRA_INCLUDES ${LLAMA_EXTRA_INCLUDES} ${BLAS_INCLUDE_DIRS})
message("${BLAS_LIBRARIES} ${BLAS_INCLUDE_DIRS}")
include_directories(${BLAS_INCLUDE_DIRS})
else()
message(WARNING "BLAS not found, please refer to "
"https://cmake.org/cmake/help/latest/module/FindBLAS.html#blas-lapack-vendors"
@ -408,7 +419,7 @@ add_library(ggml OBJECT
${GGML_SOURCES_EXTRA}
)
target_include_directories(ggml PUBLIC .)
target_include_directories(ggml PUBLIC . ${LLAMA_EXTRA_INCLUDES})
target_compile_features(ggml PUBLIC c_std_11) # don't bump
target_link_libraries(ggml PUBLIC Threads::Threads ${LLAMA_EXTRA_LIBS})

7
Makefile
View file

@ -1,5 +1,5 @@
# Define the default target now so that it is always the first target
BUILD_TARGETS = main quantize quantize-stats perplexity embedding vdot
BUILD_TARGETS = main quantize quantize-stats perplexity embedding vdot train-text-from-scratch
ifdef LLAMA_BUILD_SERVER
BUILD_TARGETS += server
@ -261,7 +261,7 @@ libllama.so: llama.o ggml.o $(OBJS)
$(CXX) $(CXXFLAGS) -shared -fPIC -o $@ $^ $(LDFLAGS)
clean:
rm -vf *.o main quantize quantize-stats perplexity embedding benchmark-matmult save-load-state server vdot build-info.h
rm -vf *.o *.so main quantize quantize-stats perplexity embedding benchmark-matmult save-load-state server vdot train-text-from-scratch build-info.h
#
# Examples
@ -291,6 +291,9 @@ save-load-state: examples/save-load-state/save-load-state.cpp build-info.h ggml.
server: examples/server/server.cpp examples/server/httplib.h examples/server/json.hpp build-info.h ggml.o llama.o common.o $(OBJS)
$(CXX) $(CXXFLAGS) -Iexamples/server $(filter-out %.h,$(filter-out %.hpp,$^)) -o $@ $(LDFLAGS)
train-text-from-scratch: examples/train-text-from-scratch/train-text-from-scratch.cpp build-info.h ggml.o llama.o $(OBJS)
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
build-info.h: $(wildcard .git/index) scripts/build-info.sh
@sh scripts/build-info.sh > $@.tmp
@if ! cmp -s $@.tmp $@; then \

1
Package.swift
View file

@ -11,6 +11,7 @@ let package = Package(
.target(
name: "llama",
path: ".",
exclude: ["ggml-metal.metal"],
sources: ["ggml.c", "llama.cpp"],
publicHeadersPath: "spm-headers",
cSettings: [.unsafeFlags(["-Wno-shorten-64-to-32"]), .define("GGML_USE_ACCELERATE")],

41
examples/chat-vicuna.sh Executable file
View file

@ -0,0 +1,41 @@
#!/bin/bash
set -e
cd "$(dirname "$0")/.." || exit
MODEL="${MODEL:-./models/ggml-vic13b-uncensored-q5_0.bin}"
PROMPT_TEMPLATE=${PROMPT_TEMPLATE:-./prompts/chat.txt}
USER_NAME="### Human"
AI_NAME="### Assistant"
# Adjust to the number of CPU cores you want to use.
N_THREAD="${N_THREAD:-8}"
# Number of tokens to predict (made it larger than default because we want a long interaction)
N_PREDICTS="${N_PREDICTS:-2048}"
# Note: you can also override the generation options by specifying them on the command line:
# For example, override the context size by doing: ./chatLLaMa --ctx_size 1024
GEN_OPTIONS="${GEN_OPTIONS:---ctx_size 2048 --temp 0.7 --top_k 40 --top_p 0.5 --repeat_last_n 256 --batch_size 1024 --repeat_penalty 1.17647}"
DATE_TIME=$(date +%H:%M)
DATE_YEAR=$(date +%Y)
PROMPT_FILE=$(mktemp -t llamacpp_prompt.XXXXXXX.txt)
sed -e "s/\[\[USER_NAME\]\]/$USER_NAME/g" \
-e "s/\[\[AI_NAME\]\]/$AI_NAME/g" \
-e "s/\[\[DATE_TIME\]\]/$DATE_TIME/g" \
-e "s/\[\[DATE_YEAR\]\]/$DATE_YEAR/g" \
$PROMPT_TEMPLATE > $PROMPT_FILE
# shellcheck disable=SC2086 # Intended splitting of GEN_OPTIONS
./bin/main $GEN_OPTIONS \
--model "$MODEL" \
--threads "$N_THREAD" \
--n_predict "$N_PREDICTS" \
--color --interactive \
--file ${PROMPT_FILE} \
--reverse-prompt "### Human:" \
--in-prefix ' ' \
"$@"

8
examples/common.cpp
View file

@ -412,6 +412,14 @@ bool gpt_params_parse(int argc, char ** argv, gpt_params & params) {
gpt_print_usage(argc, argv, default_params);
exit(1);
}
#ifdef GGML_USE_CUBLAS
if (!params.lora_adapter.empty() && params.n_gpu_layers > 0) {
fprintf(stderr, "%s: error: the simultaneous use of LoRAs and GPU acceleration is not supported", __func__);
exit(1);
}
#endif // GGML_USE_CUBLAS
if (escape_prompt) {
process_escapes(params.prompt);
}

4
ggml-cuda.cu
View file

@ -25,7 +25,7 @@ static_assert(sizeof(half) == sizeof(ggml_fp16_t), "wrong fp16 size");
} \
} while (0)
#if CUDART_VERSION >= 12
#if CUDART_VERSION >= 12000
#define CUBLAS_CHECK(err) \
do { \
cublasStatus_t err_ = (err); \
@ -2366,7 +2366,7 @@ void ggml_cuda_assign_buffers_no_scratch(struct ggml_tensor * tensor) {
}
void ggml_cuda_set_main_device(int main_device) {
if (main_device > g_device_count) {
if (main_device >= g_device_count) {
fprintf(stderr, "warning: cannot set main_device=%d because there are only %d devices. Using device %d instead.\n",
main_device, g_device_count, g_main_device);
return;

1
ggml-metal.h
View file

@ -55,6 +55,7 @@ void ggml_metal_set_tensor(struct ggml_metal_context * ctx, struct ggml_tensor *
void ggml_metal_get_tensor(struct ggml_metal_context * ctx, struct ggml_tensor * t);
// same as ggml_graph_compute but uses Metal
// creates gf->n_threads command buffers in parallel
void ggml_metal_graph_compute(struct ggml_metal_context * ctx, struct ggml_cgraph * gf);
#ifdef __cplusplus

917
ggml-metal.m
View file

@ -284,528 +284,551 @@ void ggml_metal_get_tensor(
void ggml_metal_graph_compute(
struct ggml_metal_context * ctx,
struct ggml_cgraph * gf) {
struct ggml_cgraph * gf) {
metal_printf("%s: evaluating graph\n", __func__);
size_t offs_src0 = 0;
size_t offs_src1 = 0;
size_t offs_dst = 0;
// create multiple command buffers and enqueue them
// then, we encode the graph into the command buffers in parallel
id<MTLCommandBuffer> command_buffer = [ctx->queue commandBuffer];
id<MTLComputeCommandEncoder> encoder = nil;
const int n_cb = gf->n_threads;
for (int i = 0; i < gf->n_nodes; ++i) {
//metal_printf("%s: encoding node %3d, op = %8s\n", __func__, i, ggml_op_name(gf->nodes[i]->op));
NSMutableArray * command_buffers = [NSMutableArray arrayWithCapacity:n_cb];
struct ggml_tensor * src0 = gf->nodes[i]->src0;
struct ggml_tensor * src1 = gf->nodes[i]->src1;
struct ggml_tensor * dst = gf->nodes[i];
for (int i = 0; i < n_cb; ++i) {
command_buffers[i] = [ctx->queue commandBuffer];
const int64_t ne00 = src0 ? src0->ne[0] : 0;
const int64_t ne01 = src0 ? src0->ne[1] : 0;
const int64_t ne02 = src0 ? src0->ne[2] : 0;
const int64_t ne03 = src0 ? src0->ne[3] : 0;
// enqueue the command buffers in order to specify their execution order
[command_buffers[i] enqueue];
}
const uint64_t nb00 = src0 ? src0->nb[0] : 0;
const uint64_t nb01 = src0 ? src0->nb[1] : 0;
const uint64_t nb02 = src0 ? src0->nb[2] : 0;
const uint64_t nb03 = src0 ? src0->nb[3] : 0;
// TODO: is this the best way to start threads?
dispatch_queue_t queue = dispatch_queue_create("llama.cpp", DISPATCH_QUEUE_CONCURRENT);
const int64_t ne10 = src1 ? src1->ne[0] : 0;
const int64_t ne11 = src1 ? src1->ne[1] : 0;
const int64_t ne12 = src1 ? src1->ne[2] : 0;
const int64_t ne13 = src1 ? src1->ne[3] : 0; UNUSED(ne13);
for (int cb_idx = 0; cb_idx < n_cb; ++cb_idx) {
const int n_nodes_per_cb = (gf->n_nodes + n_cb - 1) / n_cb;
const uint64_t nb10 = src1 ? src1->nb[0] : 0;
const uint64_t nb11 = src1 ? src1->nb[1] : 0;
const uint64_t nb12 = src1 ? src1->nb[2] : 0;
const uint64_t nb13 = src1 ? src1->nb[3] : 0; UNUSED(nb13);
dispatch_async(queue, ^{
size_t offs_src0 = 0;
size_t offs_src1 = 0;
size_t offs_dst = 0;
const int64_t ne0 = dst ? dst->ne[0] : 0;
const int64_t ne1 = dst ? dst->ne[1] : 0;
const int64_t ne2 = dst ? dst->ne[2] : 0;
const int64_t ne3 = dst ? dst->ne[3] : 0;
id<MTLCommandBuffer> command_buffer = command_buffers[cb_idx];
const uint64_t nb0 = dst ? dst->nb[0] : 0;
const uint64_t nb1 = dst ? dst->nb[1] : 0;
const uint64_t nb2 = dst ? dst->nb[2] : 0;
const uint64_t nb3 = dst ? dst->nb[3] : 0;
id<MTLComputeCommandEncoder> encoder = nil;
const enum ggml_type src0t = src0 ? src0->type : GGML_TYPE_COUNT;
const enum ggml_type src1t = src1 ? src1->type : GGML_TYPE_COUNT;
const enum ggml_type dstt = dst ? dst->type : GGML_TYPE_COUNT;
const int node_start = (cb_idx + 0) * n_nodes_per_cb;
const int node_end = (cb_idx == n_cb - 1) ? gf->n_nodes : (cb_idx + 1) * n_nodes_per_cb;
id<MTLBuffer> id_src0 = src0 ? ggml_metal_get_buffer(ctx, src0, &offs_src0) : nil;
id<MTLBuffer> id_src1 = src1 ? ggml_metal_get_buffer(ctx, src1, &offs_src1) : nil;
id<MTLBuffer> id_dst = dst ? ggml_metal_get_buffer(ctx, dst, &offs_dst) : nil;
for (int i = node_start; i < node_end; ++i) {
metal_printf("%s: encoding node %3d, op = %8s\n", __func__, i, ggml_op_name(gf->nodes[i]->op));
//metal_printf("%s: op - %s\n", __func__, ggml_op_name(dst->op));
//if (src0) {
// metal_printf("%s: src0 - %4s [%5lld, %5lld, %5lld], %d, %s\n", __func__, ggml_type_name(src0t), ne00, ne01, ne02,
// ggml_is_contiguous(src0), src0->name);
//}
//if (src1) {
// metal_printf("%s: src1 - %4s [%5lld, %5lld, %5lld], %d, %s\n", __func__, ggml_type_name(src1t), ne10, ne11, ne12,
// ggml_is_contiguous(src1), src1->name);
//}
//if (dst) {
// metal_printf("%s: dst - %4s [%5lld, %5lld, %5lld], 1, %s\n", __func__, ggml_type_name(dstt), ne0, ne1, ne2,
// dst->name);
//}
struct ggml_tensor * src0 = gf->nodes[i]->src0;
struct ggml_tensor * src1 = gf->nodes[i]->src1;
struct ggml_tensor * dst = gf->nodes[i];
switch (dst->op) {
case GGML_OP_RESHAPE:
case GGML_OP_VIEW:
case GGML_OP_TRANSPOSE:
case GGML_OP_PERMUTE:
{
// noop
} break;
case GGML_OP_ADD:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
const int64_t ne00 = src0 ? src0->ne[0] : 0;
const int64_t ne01 = src0 ? src0->ne[1] : 0;
const int64_t ne02 = src0 ? src0->ne[2] : 0;
const int64_t ne03 = src0 ? src0->ne[3] : 0;
[encoder setComputePipelineState:ctx->pipeline_add];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
const uint64_t nb00 = src0 ? src0->nb[0] : 0;
const uint64_t nb01 = src0 ? src0->nb[1] : 0;
const uint64_t nb02 = src0 ? src0->nb[2] : 0;
const uint64_t nb03 = src0 ? src0->nb[3] : 0;
const int64_t n = ggml_nelements(dst);
const int64_t ne10 = src1 ? src1->ne[0] : 0;
const int64_t ne11 = src1 ? src1->ne[1] : 0;
const int64_t ne12 = src1 ? src1->ne[2] : 0;
const int64_t ne13 = src1 ? src1->ne[3] : 0; UNUSED(ne13);
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_MUL:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
const uint64_t nb10 = src1 ? src1->nb[0] : 0;
const uint64_t nb11 = src1 ? src1->nb[1] : 0;
const uint64_t nb12 = src1 ? src1->nb[2] : 0;
const uint64_t nb13 = src1 ? src1->nb[3] : 0; UNUSED(nb13);
if (ggml_nelements(src1) == ne10) {
// src1 is a row
[encoder setComputePipelineState:ctx->pipeline_mul_row];
} else {
[encoder setComputePipelineState:ctx->pipeline_mul];
}
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:3];
const int64_t ne0 = dst ? dst->ne[0] : 0;
const int64_t ne1 = dst ? dst->ne[1] : 0;
const int64_t ne2 = dst ? dst->ne[2] : 0;
const int64_t ne3 = dst ? dst->ne[3] : 0;
const int64_t n = ggml_nelements(dst);
const uint64_t nb0 = dst ? dst->nb[0] : 0;
const uint64_t nb1 = dst ? dst->nb[1] : 0;
const uint64_t nb2 = dst ? dst->nb[2] : 0;
const uint64_t nb3 = dst ? dst->nb[3] : 0;
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_SCALE:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
const enum ggml_type src0t = src0 ? src0->type : GGML_TYPE_COUNT;
const enum ggml_type src1t = src1 ? src1->type : GGML_TYPE_COUNT;
const enum ggml_type dstt = dst ? dst->type : GGML_TYPE_COUNT;
const float scale = *(const float *) src1->data;
id<MTLBuffer> id_src0 = src0 ? ggml_metal_get_buffer(ctx, src0, &offs_src0) : nil;
id<MTLBuffer> id_src1 = src1 ? ggml_metal_get_buffer(ctx, src1, &offs_src1) : nil;
id<MTLBuffer> id_dst = dst ? ggml_metal_get_buffer(ctx, dst, &offs_dst) : nil;
[encoder setComputePipelineState:ctx->pipeline_scale];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&scale length:sizeof(scale) atIndex:2];
//metal_printf("%s: op - %s\n", __func__, ggml_op_name(dst->op));
//if (src0) {
// metal_printf("%s: src0 - %4s [%5lld, %5lld, %5lld], %d, %s\n", __func__, ggml_type_name(src0t), ne00, ne01, ne02,
// ggml_is_contiguous(src0), src0->name);
//}
//if (src1) {
// metal_printf("%s: src1 - %4s [%5lld, %5lld, %5lld], %d, %s\n", __func__, ggml_type_name(src1t), ne10, ne11, ne12,
// ggml_is_contiguous(src1), src1->name);
//}
//if (dst) {
// metal_printf("%s: dst - %4s [%5lld, %5lld, %5lld], 1, %s\n", __func__, ggml_type_name(dstt), ne0, ne1, ne2,
// dst->name);
//}
const int64_t n = ggml_nelements(dst);
switch (dst->op) {
case GGML_OP_RESHAPE:
case GGML_OP_VIEW:
case GGML_OP_TRANSPOSE:
case GGML_OP_PERMUTE:
{
// noop
} break;
case GGML_OP_ADD:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_SILU:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
[encoder setComputePipelineState:ctx->pipeline_add];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
[encoder setComputePipelineState:ctx->pipeline_silu];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
const int64_t n = ggml_nelements(dst);
const int64_t n = ggml_nelements(dst);
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_MUL:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_RELU:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
if (ggml_nelements(src1) == ne10) {
// src1 is a row
[encoder setComputePipelineState:ctx->pipeline_mul_row];
} else {
[encoder setComputePipelineState:ctx->pipeline_mul];
}
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:3];
[encoder setComputePipelineState:ctx->pipeline_relu];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
const int64_t n = ggml_nelements(dst);
const int64_t n = ggml_nelements(dst);
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_SCALE:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_GELU:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
const float scale = *(const float *) src1->data;
[encoder setComputePipelineState:ctx->pipeline_gelu];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setComputePipelineState:ctx->pipeline_scale];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&scale length:sizeof(scale) atIndex:2];
const int64_t n = ggml_nelements(dst);
const int64_t n = ggml_nelements(dst);
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_SOFT_MAX:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_SILU:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
const int nth = 32;
[encoder setComputePipelineState:ctx->pipeline_silu];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setComputePipelineState:ctx->pipeline_soft_max];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
[encoder setThreadgroupMemoryLength:nth*sizeof(float) atIndex:0];
const int64_t n = ggml_nelements(dst);
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
} break;
case GGML_OP_DIAG_MASK_INF:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_RELU:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
const int n_past = ((int32_t *)(src1->data))[0];
[encoder setComputePipelineState:ctx->pipeline_relu];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setComputePipelineState:ctx->pipeline_diag_mask_inf];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
[encoder setBytes:&n_past length:sizeof(int) atIndex:4];
const int64_t n = ggml_nelements(dst);
[encoder dispatchThreadgroups:MTLSizeMake(ne00, ne01, ne02) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_MUL_MAT:
{
// TODO: needs to be updated after PR: https://github.com/ggerganov/ggml/pull/224
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_GELU:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
GGML_ASSERT(ne00 == ne10);
GGML_ASSERT(ne02 == ne12);
[encoder setComputePipelineState:ctx->pipeline_gelu];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
if (ggml_is_contiguous(src0) &&
ggml_is_contiguous(src1) &&
(src0t == GGML_TYPE_F32 || src0t == GGML_TYPE_F16) && ne11 > 1) {
const int64_t n = ggml_nelements(dst);
if (encoder != nil) {
[encoder endEncoding];
encoder = nil;
}
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_SOFT_MAX:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
MPSDataType src0dt = src0t == GGML_TYPE_F32 ? MPSDataTypeFloat32 : MPSDataTypeFloat16;
MPSDataType src1dt = src1t == GGML_TYPE_F32 ? MPSDataTypeFloat32 : MPSDataTypeFloat16;
const int nth = 32;
// for F32 x F32 we use MPS
MPSMatrixDescriptor * desc0 = [MPSMatrixDescriptor
matrixDescriptorWithRows:ne01 columns:ne00 rowBytes:src0->nb[1] dataType:src0dt];
[encoder setComputePipelineState:ctx->pipeline_soft_max];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
[encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
[encoder setThreadgroupMemoryLength:nth*sizeof(float) atIndex:0];
MPSMatrixDescriptor * desc1 = [MPSMatrixDescriptor
matrixDescriptorWithRows:ne11 columns:ne10 rowBytes:src1->nb[1] dataType:src1dt];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
} break;
case GGML_OP_DIAG_MASK_INF:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
MPSMatrixDescriptor * desc = [MPSMatrixDescriptor
matrixDescriptorWithRows:ne1 columns:ne0 rowBytes:dst->nb[1] dataType:MPSDataTypeFloat32];
const int n_past = ((int32_t *)(src1->data))[0];
MPSMatrixMultiplication * mul = [[MPSMatrixMultiplication alloc]
initWithDevice:ctx->device transposeLeft:false transposeRight:true
resultRows:ne11 resultColumns:ne01 interiorColumns:ne00 alpha:1.0 beta:0.0];
[encoder setComputePipelineState:ctx->pipeline_diag_mask_inf];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
[encoder setBytes:&n_past length:sizeof(int) atIndex:4];
// we need to do ne02 multiplications
// TODO: is there a way to do this in parallel - currently very slow ..
// TODO: might be possible to offload part of the computation to ANE using Accelerate's CBLAS
for (int64_t i02 = 0; i02 < ne02; ++i02) {
size_t offs_src0_cur = offs_src0 + i02*nb02;
size_t offs_src1_cur = offs_src1 + i02*nb12;
size_t offs_dst_cur = offs_dst + i02*nb2;
[encoder dispatchThreadgroups:MTLSizeMake(ne00, ne01, ne02) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_MUL_MAT:
{
// TODO: needs to be updated after PR: https://github.com/ggerganov/ggml/pull/224
MPSMatrix * mat_src0 = [[MPSMatrix alloc] initWithBuffer:id_src0 offset:offs_src0_cur descriptor:desc0];
MPSMatrix * mat_src1 = [[MPSMatrix alloc] initWithBuffer:id_src1 offset:offs_src1_cur descriptor:desc1];
MPSMatrix * mat_dst = [[MPSMatrix alloc] initWithBuffer:id_dst offset:offs_dst_cur descriptor:desc ];
GGML_ASSERT(ne00 == ne10);
GGML_ASSERT(ne02 == ne12);
[mul encodeToCommandBuffer:command_buffer leftMatrix:mat_src1 rightMatrix:mat_src0 resultMatrix:mat_dst];
}
} else {
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
if (ggml_is_contiguous(src0) &&
ggml_is_contiguous(src1) &&
(src0t == GGML_TYPE_F32 || src0t == GGML_TYPE_F16) && ne11 > 1) {
int nth0 = 32;
int nth1 = 1;
// use custom matrix x vector kernel
switch (src0t) {
case GGML_TYPE_F16:
{
GGML_ASSERT(ne02 == ne12);
nth0 = 64;
nth1 = 1;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_f16_f32];
} break;
case GGML_TYPE_Q4_0:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
nth0 = 8;
nth1 = 8;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q4_0_f32];
} break;
case GGML_TYPE_Q4_1:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
nth0 = 8;
nth1 = 8;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q4_1_f32];
} break;
case GGML_TYPE_Q2_K:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
nth0 = 4;
nth1 = 16;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q2_k_f32];
} break;
case GGML_TYPE_Q3_K:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
nth0 = 4;
nth1 = 16;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q3_k_f32];
} break;
case GGML_TYPE_Q4_K:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
nth0 = 4;
nth1 = 16;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q4_k_f32];
} break;
case GGML_TYPE_Q5_K:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
nth0 = 4;
nth1 = 16;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q5_k_f32];
} break;
case GGML_TYPE_Q6_K:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
nth0 = 4;
nth1 = 16;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q6_k_f32];
} break;
default:
{
fprintf(stderr, "Asserting on type %d\n",(int)src0t);
GGML_ASSERT(false && "not implemented");
if (encoder != nil) {
[encoder endEncoding];
encoder = nil;
}
};
MPSDataType src0dt = src0t == GGML_TYPE_F32 ? MPSDataTypeFloat32 : MPSDataTypeFloat16;
MPSDataType src1dt = src1t == GGML_TYPE_F32 ? MPSDataTypeFloat32 : MPSDataTypeFloat16;
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:3];
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:4];
[encoder setBytes:&nb00 length:sizeof(nb00) atIndex:5];
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:6];
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:7];
[encoder setBytes:&ne10 length:sizeof(ne10) atIndex:8];
[encoder setBytes:&ne11 length:sizeof(ne11) atIndex:9];
[encoder setBytes:&nb10 length:sizeof(nb10) atIndex:10];
[encoder setBytes:&nb11 length:sizeof(nb11) atIndex:11];
[encoder setBytes:&nb12 length:sizeof(nb12) atIndex:12];
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:13];
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:14];
// for F32 x F32 we use MPS
MPSMatrixDescriptor * desc0 = [MPSMatrixDescriptor
matrixDescriptorWithRows:ne01 columns:ne00 rowBytes:src0->nb[1] dataType:src0dt];
if (src0t == GGML_TYPE_Q4_0 || src0t == GGML_TYPE_Q4_1) {
[encoder setThreadgroupMemoryLength:nth0*nth1*sizeof(float) atIndex:0];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne11, 1) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
}
else if (src0t == GGML_TYPE_Q2_K ||
src0t == GGML_TYPE_Q3_K ||
src0t == GGML_TYPE_Q4_K ||
src0t == GGML_TYPE_Q5_K ||
src0t == GGML_TYPE_Q6_K) {
[encoder setThreadgroupMemoryLength:nth0*nth1*sizeof(float) atIndex:0];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
} else {
[encoder setThreadgroupMemoryLength:nth0*sizeof(float) atIndex:0];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne11, ne12) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
}
}
} break;
case GGML_OP_GET_ROWS:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
MPSMatrixDescriptor * desc1 = [MPSMatrixDescriptor
matrixDescriptorWithRows:ne11 columns:ne10 rowBytes:src1->nb[1] dataType:src1dt];
switch (src0->type) {
case GGML_TYPE_F16: [encoder setComputePipelineState:ctx->pipeline_get_rows_f16]; break;
case GGML_TYPE_Q4_0: [encoder setComputePipelineState:ctx->pipeline_get_rows_q4_0]; break;
case GGML_TYPE_Q4_1: [encoder setComputePipelineState:ctx->pipeline_get_rows_q4_1]; break;
case GGML_TYPE_Q2_K: [encoder setComputePipelineState:ctx->pipeline_get_rows_q2_k]; break;
case GGML_TYPE_Q3_K: [encoder setComputePipelineState:ctx->pipeline_get_rows_q3_k]; break;
case GGML_TYPE_Q4_K: [encoder setComputePipelineState:ctx->pipeline_get_rows_q4_k]; break;
case GGML_TYPE_Q5_K: [encoder setComputePipelineState:ctx->pipeline_get_rows_q5_k]; break;
case GGML_TYPE_Q6_K: [encoder setComputePipelineState:ctx->pipeline_get_rows_q6_k]; break;
default: GGML_ASSERT(false && "not implemented");
}
MPSMatrixDescriptor * desc = [MPSMatrixDescriptor
matrixDescriptorWithRows:ne1 columns:ne0 rowBytes:dst->nb[1] dataType:MPSDataTypeFloat32];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
[encoder setBytes:&(src0->ne[0]) length:sizeof( int64_t) atIndex:3];
[encoder setBytes:&(src0->nb[1]) length:sizeof(uint64_t) atIndex:4];
[encoder setBytes:&(dst->nb[1]) length:sizeof(uint64_t) atIndex:5];
MPSMatrixMultiplication * mul = [[MPSMatrixMultiplication alloc]
initWithDevice:ctx->device transposeLeft:false transposeRight:true
resultRows:ne11 resultColumns:ne01 interiorColumns:ne00 alpha:1.0 beta:0.0];
const int64_t n = ggml_nelements(src1);
// we need to do ne02 multiplications
// TODO: is there a way to do this in parallel - currently very slow ..
// TODO: might be possible to offload part of the computation to ANE using Accelerate's CBLAS
for (int64_t i02 = 0; i02 < ne02; ++i02) {
size_t offs_src0_cur = offs_src0 + i02*nb02;
size_t offs_src1_cur = offs_src1 + i02*nb12;
size_t offs_dst_cur = offs_dst + i02*nb2;
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_RMS_NORM:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
MPSMatrix * mat_src0 = [[MPSMatrix alloc] initWithBuffer:id_src0 offset:offs_src0_cur descriptor:desc0];
MPSMatrix * mat_src1 = [[MPSMatrix alloc] initWithBuffer:id_src1 offset:offs_src1_cur descriptor:desc1];
MPSMatrix * mat_dst = [[MPSMatrix alloc] initWithBuffer:id_dst offset:offs_dst_cur descriptor:desc ];
const float eps = 1e-6f;
[mul encodeToCommandBuffer:command_buffer leftMatrix:mat_src1 rightMatrix:mat_src0 resultMatrix:mat_dst];
}
} else {
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
const int nth = 256;
int nth0 = 32;
int nth1 = 1;
[encoder setComputePipelineState:ctx->pipeline_rms_norm];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:3];
[encoder setBytes:&eps length:sizeof( float) atIndex:4];
[encoder setThreadgroupMemoryLength:nth*sizeof(float) atIndex:0];
// use custom matrix x vector kernel
switch (src0t) {
case GGML_TYPE_F16:
{
GGML_ASSERT(ne02 == ne12);
const int64_t nrows = ggml_nrows(src0);
nth0 = 64;
nth1 = 1;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_f16_f32];
} break;
case GGML_TYPE_Q4_0:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
[encoder dispatchThreadgroups:MTLSizeMake(nrows, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
} break;
case GGML_OP_ROPE:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
nth0 = 8;
nth1 = 8;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q4_0_f32];
} break;
case GGML_TYPE_Q4_1:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
const int n_dims = ((int32_t *) src1->data)[1];
const int mode = ((int32_t *) src1->data)[2];
nth0 = 8;
nth1 = 8;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q4_1_f32];
} break;
case GGML_TYPE_Q2_K:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
const int n_past = ((int32_t *)(src1->data))[0];
nth0 = 4;
nth1 = 16;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q2_k_f32];
} break;
case GGML_TYPE_Q3_K:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
[encoder setComputePipelineState:ctx->pipeline_rope];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
[encoder setBytes:&ne01 length:sizeof( int64_t) atIndex:3];
[encoder setBytes:&ne02 length:sizeof( int64_t) atIndex:4];
[encoder setBytes:&ne03 length:sizeof( int64_t) atIndex:5];
[encoder setBytes:&nb00 length:sizeof(uint64_t) atIndex:6];
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:7];
[encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:8];
[encoder setBytes:&nb03 length:sizeof(uint64_t) atIndex:9];
[encoder setBytes:&ne0 length:sizeof( int64_t) atIndex:10];
[encoder setBytes:&ne1 length:sizeof( int64_t) atIndex:11];
[encoder setBytes:&ne2 length:sizeof( int64_t) atIndex:12];
[encoder setBytes:&ne3 length:sizeof( int64_t) atIndex:13];
[encoder setBytes:&nb0 length:sizeof(uint64_t) atIndex:14];
[encoder setBytes:&nb1 length:sizeof(uint64_t) atIndex:15];
[encoder setBytes:&nb2 length:sizeof(uint64_t) atIndex:16];
[encoder setBytes:&nb3 length:sizeof(uint64_t) atIndex:17];
[encoder setBytes:&n_past length:sizeof( int) atIndex:18];
[encoder setBytes:&n_dims length:sizeof( int) atIndex:19];
[encoder setBytes:&mode length:sizeof( int) atIndex:20];
nth0 = 4;
nth1 = 16;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q3_k_f32];
} break;
case GGML_TYPE_Q4_K:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_CPY:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
nth0 = 4;
nth1 = 16;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q4_k_f32];
} break;
case GGML_TYPE_Q5_K:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
const int nth = 32;
nth0 = 4;
nth1 = 16;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q5_k_f32];
} break;
case GGML_TYPE_Q6_K:
{
GGML_ASSERT(ne02 == 1);
GGML_ASSERT(ne12 == 1);
switch (src0t) {
case GGML_TYPE_F32:
{
switch (dstt) {
case GGML_TYPE_F16: [encoder setComputePipelineState:ctx->pipeline_cpy_f32_f16]; break;
case GGML_TYPE_F32: [encoder setComputePipelineState:ctx->pipeline_cpy_f32_f32]; break;
default: GGML_ASSERT(false && "not implemented");
nth0 = 4;
nth1 = 16;
[encoder setComputePipelineState:ctx->pipeline_mul_mat_q6_k_f32];
} break;
default:
{
fprintf(stderr, "Asserting on type %d\n",(int)src0t);
GGML_ASSERT(false && "not implemented");
}
};
} break;
default: GGML_ASSERT(false && "not implemented");
}
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
[encoder setBytes:&ne01 length:sizeof( int64_t) atIndex:3];
[encoder setBytes:&ne02 length:sizeof( int64_t) atIndex:4];
[encoder setBytes:&ne03 length:sizeof( int64_t) atIndex:5];
[encoder setBytes:&nb00 length:sizeof(uint64_t) atIndex:6];
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:7];
[encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:8];
[encoder setBytes:&nb03 length:sizeof(uint64_t) atIndex:9];
[encoder setBytes:&ne0 length:sizeof( int64_t) atIndex:10];
[encoder setBytes:&ne1 length:sizeof( int64_t) atIndex:11];
[encoder setBytes:&ne2 length:sizeof( int64_t) atIndex:12];
[encoder setBytes:&ne3 length:sizeof( int64_t) atIndex:13];
[encoder setBytes:&nb0 length:sizeof(uint64_t) atIndex:14];
[encoder setBytes:&nb1 length:sizeof(uint64_t) atIndex:15];
[encoder setBytes:&nb2 length:sizeof(uint64_t) atIndex:16];
[encoder setBytes:&nb3 length:sizeof(uint64_t) atIndex:17];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
[encoder setBytes:&ne00 length:sizeof(ne00) atIndex:3];
[encoder setBytes:&ne01 length:sizeof(ne01) atIndex:4];
[encoder setBytes:&nb00 length:sizeof(nb00) atIndex:5];
[encoder setBytes:&nb01 length:sizeof(nb01) atIndex:6];
[encoder setBytes:&nb02 length:sizeof(nb02) atIndex:7];
[encoder setBytes:&ne10 length:sizeof(ne10) atIndex:8];
[encoder setBytes:&ne11 length:sizeof(ne11) atIndex:9];
[encoder setBytes:&nb10 length:sizeof(nb10) atIndex:10];
[encoder setBytes:&nb11 length:sizeof(nb11) atIndex:11];
[encoder setBytes:&nb12 length:sizeof(nb12) atIndex:12];
[encoder setBytes:&ne0 length:sizeof(ne0) atIndex:13];
[encoder setBytes:&ne1 length:sizeof(ne1) atIndex:14];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
} break;
default:
fprintf(stderr, "%s: node %3d, op = %8s not implemented\n", __func__, i, ggml_op_name(dst->op));
GGML_ASSERT(false);
}
if (src0t == GGML_TYPE_Q4_0 || src0t == GGML_TYPE_Q4_1) {
[encoder setThreadgroupMemoryLength:nth0*nth1*sizeof(float) atIndex:0];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne11, 1) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
}
else if (src0t == GGML_TYPE_Q2_K ||
src0t == GGML_TYPE_Q3_K ||
src0t == GGML_TYPE_Q4_K ||
src0t == GGML_TYPE_Q5_K ||
src0t == GGML_TYPE_Q6_K) {
[encoder setThreadgroupMemoryLength:nth0*nth1*sizeof(float) atIndex:0];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
} else {
[encoder setThreadgroupMemoryLength:nth0*sizeof(float) atIndex:0];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne11, ne12) threadsPerThreadgroup:MTLSizeMake(nth0, nth1, 1)];
}
}
} break;
case GGML_OP_GET_ROWS:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
switch (src0->type) {
case GGML_TYPE_F16: [encoder setComputePipelineState:ctx->pipeline_get_rows_f16]; break;
case GGML_TYPE_Q4_0: [encoder setComputePipelineState:ctx->pipeline_get_rows_q4_0]; break;
case GGML_TYPE_Q4_1: [encoder setComputePipelineState:ctx->pipeline_get_rows_q4_1]; break;
case GGML_TYPE_Q2_K: [encoder setComputePipelineState:ctx->pipeline_get_rows_q2_k]; break;
case GGML_TYPE_Q3_K: [encoder setComputePipelineState:ctx->pipeline_get_rows_q3_k]; break;
case GGML_TYPE_Q4_K: [encoder setComputePipelineState:ctx->pipeline_get_rows_q4_k]; break;
case GGML_TYPE_Q5_K: [encoder setComputePipelineState:ctx->pipeline_get_rows_q5_k]; break;
case GGML_TYPE_Q6_K: [encoder setComputePipelineState:ctx->pipeline_get_rows_q6_k]; break;
default: GGML_ASSERT(false && "not implemented");
}
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
[encoder setBuffer:id_dst offset:offs_dst atIndex:2];
[encoder setBytes:&(src0->ne[0]) length:sizeof( int64_t) atIndex:3];
[encoder setBytes:&(src0->nb[1]) length:sizeof(uint64_t) atIndex:4];
[encoder setBytes:&(dst->nb[1]) length:sizeof(uint64_t) atIndex:5];
const int64_t n = ggml_nelements(src1);
[encoder dispatchThreadgroups:MTLSizeMake(n, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_RMS_NORM:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
const float eps = 1e-6f;
const int nth = 256;
[encoder setComputePipelineState:ctx->pipeline_rms_norm];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:3];
[encoder setBytes:&eps length:sizeof( float) atIndex:4];
[encoder setThreadgroupMemoryLength:nth*sizeof(float) atIndex:0];
const int64_t nrows = ggml_nrows(src0);
[encoder dispatchThreadgroups:MTLSizeMake(nrows, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
} break;
case GGML_OP_ROPE:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
const int n_dims = ((int32_t *) src1->data)[1];
const int mode = ((int32_t *) src1->data)[2];
const int n_past = ((int32_t *)(src1->data))[0];
[encoder setComputePipelineState:ctx->pipeline_rope];
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
[encoder setBytes:&ne01 length:sizeof( int64_t) atIndex:3];
[encoder setBytes:&ne02 length:sizeof( int64_t) atIndex:4];
[encoder setBytes:&ne03 length:sizeof( int64_t) atIndex:5];
[encoder setBytes:&nb00 length:sizeof(uint64_t) atIndex:6];
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:7];
[encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:8];
[encoder setBytes:&nb03 length:sizeof(uint64_t) atIndex:9];
[encoder setBytes:&ne0 length:sizeof( int64_t) atIndex:10];
[encoder setBytes:&ne1 length:sizeof( int64_t) atIndex:11];
[encoder setBytes:&ne2 length:sizeof( int64_t) atIndex:12];
[encoder setBytes:&ne3 length:sizeof( int64_t) atIndex:13];
[encoder setBytes:&nb0 length:sizeof(uint64_t) atIndex:14];
[encoder setBytes:&nb1 length:sizeof(uint64_t) atIndex:15];
[encoder setBytes:&nb2 length:sizeof(uint64_t) atIndex:16];
[encoder setBytes:&nb3 length:sizeof(uint64_t) atIndex:17];
[encoder setBytes:&n_past length:sizeof( int) atIndex:18];
[encoder setBytes:&n_dims length:sizeof( int) atIndex:19];
[encoder setBytes:&mode length:sizeof( int) atIndex:20];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
} break;
case GGML_OP_CPY:
{
if (encoder == nil) {
encoder = [command_buffer computeCommandEncoder];
}
const int nth = 32;
switch (src0t) {
case GGML_TYPE_F32:
{
switch (dstt) {
case GGML_TYPE_F16: [encoder setComputePipelineState:ctx->pipeline_cpy_f32_f16]; break;
case GGML_TYPE_F32: [encoder setComputePipelineState:ctx->pipeline_cpy_f32_f32]; break;
default: GGML_ASSERT(false && "not implemented");
};
} break;
default: GGML_ASSERT(false && "not implemented");
}
[encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
[encoder setBuffer:id_dst offset:offs_dst atIndex:1];
[encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:2];
[encoder setBytes:&ne01 length:sizeof( int64_t) atIndex:3];
[encoder setBytes:&ne02 length:sizeof( int64_t) atIndex:4];
[encoder setBytes:&ne03 length:sizeof( int64_t) atIndex:5];
[encoder setBytes:&nb00 length:sizeof(uint64_t) atIndex:6];
[encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:7];
[encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:8];
[encoder setBytes:&nb03 length:sizeof(uint64_t) atIndex:9];
[encoder setBytes:&ne0 length:sizeof( int64_t) atIndex:10];
[encoder setBytes:&ne1 length:sizeof( int64_t) atIndex:11];
[encoder setBytes:&ne2 length:sizeof( int64_t) atIndex:12];
[encoder setBytes:&ne3 length:sizeof( int64_t) atIndex:13];
[encoder setBytes:&nb0 length:sizeof(uint64_t) atIndex:14];
[encoder setBytes:&nb1 length:sizeof(uint64_t) atIndex:15];
[encoder setBytes:&nb2 length:sizeof(uint64_t) atIndex:16];
[encoder setBytes:&nb3 length:sizeof(uint64_t) atIndex:17];
[encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
} break;
default:
fprintf(stderr, "%s: node %3d, op = %8s not implemented\n", __func__, i, ggml_op_name(dst->op));
GGML_ASSERT(false);
}
}
if (encoder != nil) {
[encoder endEncoding];
encoder = nil;
}
[command_buffer commit];
});
}
if (encoder != nil) {
[encoder endEncoding];
encoder = nil;
}
// wait for all threads to finish
dispatch_barrier_sync(queue, ^{});
[command_buffer commit];
[command_buffer waitUntilCompleted];
{
const double time_elapsed = [command_buffer GPUEndTime] - [command_buffer GPUStartTime];
UNUSED(time_elapsed);
metal_printf("%s: time elapsed = %f ms\n", __func__, time_elapsed * 1000.0);
}
[command_buffers[n_cb - 1] waitUntilCompleted];
}

6
llama.h
View file

@ -244,9 +244,9 @@ extern "C" {
LLAMA_API const char * llama_token_to_str(const struct llama_context * ctx, llama_token token);
// Special tokens
LLAMA_API llama_token llama_token_bos();
LLAMA_API llama_token llama_token_eos();
LLAMA_API llama_token llama_token_nl();
LLAMA_API llama_token llama_token_bos(); // beginning-of-sentence
LLAMA_API llama_token llama_token_eos(); // end-of-sentence
LLAMA_API llama_token llama_token_nl(); // next-line
// Sampling functions

1
spm-headers/ggml.h Symbolic link
View file

@ -0,0 +1 @@
../ggml.h