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

Merge remote-tracking branch 'origin/master' into sl/ggml-backend-int

Browse source
This commit is contained in:
slaren 2023-12-19 03:52:34 +01:00
commit 94507911bb
11 changed files with 485 additions and 79 deletions
Download patch file Download diff file Expand all files Collapse all files

22
convert-hf-to-gguf.py
View file

@ -182,6 +182,8 @@ class Model:
return QwenModel return QwenModel
if model_architecture == "MixtralForCausalLM": if model_architecture == "MixtralForCausalLM":
return MixtralModel return MixtralModel
if model_architecture == "PhiForCausalLM":
return Phi2Model
return Model return Model
def _is_model_safetensors(self) -> bool: def _is_model_safetensors(self) -> bool:
@ -221,6 +223,8 @@ class Model:
return gguf.MODEL_ARCH.QWEN return gguf.MODEL_ARCH.QWEN
if arch == "MixtralForCausalLM": if arch == "MixtralForCausalLM":
return gguf.MODEL_ARCH.LLAMA return gguf.MODEL_ARCH.LLAMA
if arch == "PhiForCausalLM":
return gguf.MODEL_ARCH.PHI2
raise NotImplementedError(f'Architecture "{arch}" not supported!') raise NotImplementedError(f'Architecture "{arch}" not supported!')
@ -980,6 +984,24 @@ class QwenModel(Model):
print(f"{new_name}, n_dims = {n_dims}, {old_dtype} --> {data.dtype}") print(f"{new_name}, n_dims = {n_dims}, {old_dtype} --> {data.dtype}")
self.gguf_writer.add_tensor(new_name, data) self.gguf_writer.add_tensor(new_name, data)
class Phi2Model(Model):
def set_gguf_parameters(self):
block_count = self.hparams["n_layer"]
self.gguf_writer.add_name("Phi2")
self.gguf_writer.add_context_length(self.hparams["n_positions"])
self.gguf_writer.add_embedding_length(self.hparams["n_embd"])
self.gguf_writer.add_feed_forward_length(4 * self.hparams["n_embd"])
self.gguf_writer.add_block_count(block_count)
self.gguf_writer.add_head_count(self.hparams["n_head"])
self.gguf_writer.add_head_count_kv(self.hparams["n_head"])
self.gguf_writer.add_layer_norm_eps(self.hparams["layer_norm_epsilon"])
self.gguf_writer.add_rope_dimension_count(self.hparams["rotary_dim"])
self.gguf_writer.add_file_type(self.ftype)
self.gguf_writer.add_add_bos_token(False)
###### CONVERSION LOGIC ###### ###### CONVERSION LOGIC ######

2
examples/llama.swiftui/llama.cpp.swift/LibLlama.swift
View file

@ -203,7 +203,7 @@ actor LlamaContext {
var pp_std: Double = 0 var pp_std: Double = 0
var tg_std: Double = 0 var tg_std: Double = 0
for r in 0..<nr { for _ in 0..<nr {
// bench prompt processing // bench prompt processing
llama_batch_clear(&batch) llama_batch_clear(&batch)

39
examples/llama.swiftui/llama.swiftui/UI/ContentView.swift
View file

@ -75,21 +75,56 @@ struct ContentView: View {
VStack { VStack {
DownloadButton( DownloadButton(
llamaState: llamaState, llamaState: llamaState,
modelName: "TinyLlama-1.1B (Q4_0)", modelName: "TinyLlama-1.1B (Q4_0, 0.6 GiB)",
modelUrl: "https://huggingface.co/TheBloke/TinyLlama-1.1B-1T-OpenOrca-GGUF/resolve/main/tinyllama-1.1b-1t-openorca.Q4_0.gguf?download=true", modelUrl: "https://huggingface.co/TheBloke/TinyLlama-1.1B-1T-OpenOrca-GGUF/resolve/main/tinyllama-1.1b-1t-openorca.Q4_0.gguf?download=true",
filename: "tinyllama-1.1b-1t-openorca.Q4_0.gguf" filename: "tinyllama-1.1b-1t-openorca.Q4_0.gguf"
) )
.font(.system(size: 12)) .font(.system(size: 12))
.padding(.top, 4) .padding(.top, 4)
.frame(maxWidth: .infinity, alignment: .leading)
DownloadButton( DownloadButton(
llamaState: llamaState, llamaState: llamaState,
modelName: "TinyLlama-1.1B (Q8_0)", modelName: "TinyLlama-1.1B (Q8_0, 1.1 GiB)",
modelUrl: "https://huggingface.co/TheBloke/TinyLlama-1.1B-1T-OpenOrca-GGUF/resolve/main/tinyllama-1.1b-1t-openorca.Q8_0.gguf?download=true", modelUrl: "https://huggingface.co/TheBloke/TinyLlama-1.1B-1T-OpenOrca-GGUF/resolve/main/tinyllama-1.1b-1t-openorca.Q8_0.gguf?download=true",
filename: "tinyllama-1.1b-1t-openorca.Q8_0.gguf" filename: "tinyllama-1.1b-1t-openorca.Q8_0.gguf"
) )
.font(.system(size: 12)) .font(.system(size: 12))
DownloadButton(
llamaState: llamaState,
modelName: "TinyLlama-1.1B (F16, 2.2 GiB)",
modelUrl: "https://huggingface.co/ggml-org/models/resolve/main/tinyllama-1.1b/ggml-model-f16.gguf?download=true",
filename: "tinyllama-1.1b-f16.gguf"
)
.font(.system(size: 12))
.frame(maxWidth: .infinity, alignment: .leading)
DownloadButton(
llamaState: llamaState,
modelName: "Phi-2.7B (Q4_0, 1.6 GiB)",
modelUrl: "https://huggingface.co/ggml-org/models/resolve/main/phi-2/ggml-model-q4_0.gguf?download=true",
filename: "phi-2-q4_0.gguf"
)
.font(.system(size: 12))
DownloadButton(
llamaState: llamaState,
modelName: "Phi-2.7B (Q8_0, 2.8 GiB)",
modelUrl: "https://huggingface.co/ggml-org/models/resolve/main/phi-2/ggml-model-q8_0.gguf?download=true",
filename: "phi-2-q8_0.gguf"
)
.font(.system(size: 12))
.frame(maxWidth: .infinity, alignment: .leading)
DownloadButton(
llamaState: llamaState,
modelName: "Mistral-7B-v0.1 (Q4_0, 3.8 GiB)",
modelUrl: "https://huggingface.co/TheBloke/Mistral-7B-v0.1-GGUF/resolve/main/mistral-7b-v0.1.Q4_0.gguf?download=true",
filename: "mistral-7b-v0.1.Q4_0.gguf"
)
.font(.system(size: 12))
Button("Clear downloaded models") { Button("Clear downloaded models") {
ContentView.cleanupModelCaches() ContentView.cleanupModelCaches()
llamaState.cacheCleared = true llamaState.cacheCleared = true

105
ggml-cuda.cu
View file

@ -31,6 +31,7 @@
#define CUDA_R_16F HIPBLAS_R_16F #define CUDA_R_16F HIPBLAS_R_16F
#define CUDA_R_32F HIPBLAS_R_32F #define CUDA_R_32F HIPBLAS_R_32F
#define __shfl_xor_sync(mask, var, laneMask, width) __shfl_xor(var, laneMask, width) #define __shfl_xor_sync(mask, var, laneMask, width) __shfl_xor(var, laneMask, width)
#define cublasComputeType_t hipblasDatatype_t //deprecated, new hipblasComputeType_t not in 5.6
#define cublasCreate hipblasCreate #define cublasCreate hipblasCreate
#define cublasGemmEx hipblasGemmEx #define cublasGemmEx hipblasGemmEx
#define cublasGemmBatchedEx hipblasGemmBatchedEx #define cublasGemmBatchedEx hipblasGemmBatchedEx
@ -40,6 +41,7 @@
#define cublasSetStream hipblasSetStream #define cublasSetStream hipblasSetStream
#define cublasSgemm hipblasSgemm #define cublasSgemm hipblasSgemm
#define cublasStatus_t hipblasStatus_t #define cublasStatus_t hipblasStatus_t
#define cudaDataType_t hipblasDatatype_t //deprecated, new hipblasDatatype not in 5.6
#define cudaDeviceCanAccessPeer hipDeviceCanAccessPeer #define cudaDeviceCanAccessPeer hipDeviceCanAccessPeer
#define cudaDeviceDisablePeerAccess hipDeviceDisablePeerAccess #define cudaDeviceDisablePeerAccess hipDeviceDisablePeerAccess
#define cudaDeviceEnablePeerAccess hipDeviceEnablePeerAccess #define cudaDeviceEnablePeerAccess hipDeviceEnablePeerAccess
@ -4998,6 +5000,15 @@ static __global__ void rope_neox(
const int ib = col / n_dims; const int ib = col / n_dims;
const int ic = col % n_dims; const int ic = col % n_dims;
if (ib > 0) {
const int i = row*ncols + ib*n_dims + ic;
dst[i + 0] = x[i + 0];
dst[i + 1] = x[i + 1];
return;
}
const int i = row*ncols + ib*n_dims + ic/2; const int i = row*ncols + ib*n_dims + ic/2;
const int i2 = row/p_delta_rows; const int i2 = row/p_delta_rows;
@ -7378,7 +7389,7 @@ inline void ggml_cuda_op_mul_mat_cublas(
const int compute_capability = g_compute_capabilities[id]; const int compute_capability = g_compute_capabilities[id];
if (compute_capability >= CC_VOLTA && (src0->type == GGML_TYPE_F16 || ggml_is_quantized(src0->type)) && ggml_is_contiguous(src0) && row_diff == src0->ne[1]) { if (compute_capability >= CC_VOLTA && (src0->type == GGML_TYPE_F16 || ggml_is_quantized(src0->type)) && ggml_is_contiguous(src0) && row_diff == src0->ne[1] && dst->op_params[0] == GGML_PREC_DEFAULT) {
// convert src0 and src1 to fp16, multiply as fp16, convert dst to fp32 // convert src0 and src1 to fp16, multiply as fp16, convert dst to fp32
half * src0_as_f16 = nullptr; half * src0_as_f16 = nullptr;
size_t src0_as = 0; size_t src0_as = 0;
@ -8302,27 +8313,27 @@ static void ggml_cuda_mul_mat_vec_nc(const ggml_tensor * src0, const ggml_tensor
} }
static __global__ void k_compute_batched_ptrs( static __global__ void k_compute_batched_ptrs(
const half * src0_as_f16, const half * src1_as_f16, half * dst_f16, const half * src0_as_f16, const half * src1_as_f16, char * dst,
const void ** ptrs_src, void ** ptrs_dst, const void ** ptrs_src, void ** ptrs_dst,
int ne12, int ne13, int64_t ne12, int64_t ne13,
int ne23, int64_t ne23,
int nb02, int nb03, size_t nb02, size_t nb03,
int nb12, int nb13, size_t nb12, size_t nb13,
int nb2, int nb3, size_t nbd2, size_t nbd3,
int r2, int r3) { int64_t r2, int64_t r3) {
int i13 = blockIdx.x * blockDim.x + threadIdx.x; int64_t i13 = blockIdx.x * blockDim.x + threadIdx.x;
int i12 = blockIdx.y * blockDim.y + threadIdx.y; int64_t i12 = blockIdx.y * blockDim.y + threadIdx.y;
if (i13 >= ne13 || i12 >= ne12) { if (i13 >= ne13 || i12 >= ne12) {
return; return;
} }
int i03 = i13 / r3; int64_t i03 = i13 / r3;
int i02 = i12 / r2; int64_t i02 = i12 / r2;
ptrs_src[0*ne23 + i12 + i13*ne12] = (const char *) src0_as_f16 + i02*nb02 + i03*nb03; ptrs_src[0*ne23 + i12 + i13*ne12] = (const char *) src0_as_f16 + i02*nb02 + i03*nb03;
ptrs_src[1*ne23 + i12 + i13*ne12] = (const char *) src1_as_f16 + i12*nb12/2 + i13*nb13/2; ptrs_src[1*ne23 + i12 + i13*ne12] = (const char *) src1_as_f16 + i12*nb12/2 + i13*nb13/2;
ptrs_dst[0*ne23 + i12 + i13*ne12] = ( char *) dst_f16 + i12* nb2/2 + i13* nb3/2; ptrs_dst[0*ne23 + i12 + i13*ne12] = ( char *) dst + i12*nbd2 + i13*nbd3;
} }
static void ggml_cuda_mul_mat_mat_batched_cublas(const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) { static void ggml_cuda_mul_mat_mat_batched_cublas(const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
@ -8378,7 +8389,41 @@ static void ggml_cuda_mul_mat_mat_batched_cublas(const ggml_tensor * src0, const
to_fp16_cuda(src1_ddf, src1_as_f16, ne1, main_stream); to_fp16_cuda(src1_ddf, src1_as_f16, ne1, main_stream);
size_t dst_as = 0; size_t dst_as = 0;
half * dst_f16 = (half *) ggml_cuda_pool_malloc(ne * sizeof(half), &dst_as);
half * dst_f16 = nullptr;
char * dst_t = nullptr;
cublasComputeType_t cu_compute_type = CUBLAS_COMPUTE_16F;
cudaDataType_t cu_data_type = CUDA_R_16F;
// dst strides
size_t nbd2 = dst->nb[2];
size_t nbd3 = dst->nb[3];
const half alpha_f16 = 1.0f;
const half beta_f16 = 0.0f;
const float alpha_f32 = 1.0f;
const float beta_f32 = 0.0f;
const void * alpha = &alpha_f16;
const void * beta = &beta_f16;
if (dst->op_params[0] == GGML_PREC_DEFAULT) {
dst_f16 = (half *) ggml_cuda_pool_malloc(ne * sizeof(half), &dst_as);
dst_t = (char *) dst_f16;
nbd2 /= sizeof(float) / sizeof(half);
nbd3 /= sizeof(float) / sizeof(half);
} else {
dst_t = (char *) dst_ddf;
cu_compute_type = CUBLAS_COMPUTE_32F;
cu_data_type = CUDA_R_32F;
alpha = &alpha_f32;
beta = &beta_f32;
}
GGML_ASSERT(ne12 % ne02 == 0); GGML_ASSERT(ne12 % ne02 == 0);
GGML_ASSERT(ne13 % ne03 == 0); GGML_ASSERT(ne13 % ne03 == 0);
@ -8387,9 +8432,6 @@ static void ggml_cuda_mul_mat_mat_batched_cublas(const ggml_tensor * src0, const
const int64_t r2 = ne12/ne02; const int64_t r2 = ne12/ne02;
const int64_t r3 = ne13/ne03; const int64_t r3 = ne13/ne03;
const half alpha_f16 = 1.0f;
const half beta_f16 = 0.0f;
#if 0 #if 0
// use cublasGemmEx // use cublasGemmEx
{ {
@ -8399,12 +8441,12 @@ static void ggml_cuda_mul_mat_mat_batched_cublas(const ggml_tensor * src0, const
int i02 = i12 / r2; int i02 = i12 / r2;
CUBLAS_CHECK( CUBLAS_CHECK(
cublasGemmEx(g_cublas_handles[id], CUBLAS_OP_T, CUBLAS_OP_N, cublasGemmEx(g_cublas_handles[g_main_device], CUBLAS_OP_T, CUBLAS_OP_N,
ne01, ne11, ne10, ne01, ne11, ne10,
&alpha_f16, (const char *) src0_as_f16 + i02*src0->nb[2] + i03*src0->nb[3] , CUDA_R_16F, nb01/sizeof(half), alpha, (const char *) src0_as_f16 + i02*src0->nb[2] + i03*src0->nb[3] , CUDA_R_16F, nb01/sizeof(half),
(const char *) src1_as_f16 + i12*src1->nb[2]/2 + i13*src1->nb[3]/2, CUDA_R_16F, nb11/sizeof(float), (const char *) src1_as_f16 + i12*src1->nb[2]/2 + i13*src1->nb[3]/2, CUDA_R_16F, nb11/sizeof(float),
&beta_f16, ( char *) dst_f16 + i12* dst->nb[2]/2 + i13* dst->nb[3]/2, CUDA_R_16F, ne01, beta, ( char *) dst_t + i12*nbd2 + i13*nbd3, cu_data_type, ne01,
CUBLAS_COMPUTE_16F, cu_compute_type,
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); CUBLAS_GEMM_DEFAULT_TENSOR_OP));
} }
} }
@ -8416,11 +8458,11 @@ static void ggml_cuda_mul_mat_mat_batched_cublas(const ggml_tensor * src0, const
CUBLAS_CHECK( CUBLAS_CHECK(
cublasGemmStridedBatchedEx(g_cublas_handles[g_main_device], CUBLAS_OP_T, CUBLAS_OP_N, cublasGemmStridedBatchedEx(g_cublas_handles[g_main_device], CUBLAS_OP_T, CUBLAS_OP_N,
ne01, ne11, ne10, ne01, ne11, ne10,
&alpha_f16, (const char *) src0_as_f16, CUDA_R_16F, nb01/sizeof(half), src0->nb[2]/sizeof(half), // strideA alpha, (const char *) src0_as_f16, CUDA_R_16F, nb01/sizeof(half), src0->nb[2]/sizeof(half), // strideA
(const char *) src1_as_f16, CUDA_R_16F, nb11/sizeof(float), src1->nb[2]/sizeof(float), // strideB (const char *) src1_as_f16, CUDA_R_16F, nb11/sizeof(float), src1->nb[2]/sizeof(float), // strideB
&beta_f16, ( char *) dst_f16, CUDA_R_16F, ne01, dst->nb[2]/sizeof(float), // strideC beta, ( char *) dst_t, cu_data_type, ne01, dst->nb[2]/sizeof(float), // strideC
ne12*ne13, ne12*ne13,
CUBLAS_COMPUTE_16F, cu_compute_type,
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); CUBLAS_GEMM_DEFAULT_TENSOR_OP));
} else { } else {
// use cublasGemmBatchedEx // use cublasGemmBatchedEx
@ -8437,24 +8479,24 @@ static void ggml_cuda_mul_mat_mat_batched_cublas(const ggml_tensor * src0, const
dim3 block_dims(ne13, ne12); dim3 block_dims(ne13, ne12);
k_compute_batched_ptrs<<<1, block_dims, 0, main_stream>>>( k_compute_batched_ptrs<<<1, block_dims, 0, main_stream>>>(
src0_as_f16, src1_as_f16, dst_f16, src0_as_f16, src1_as_f16, dst_t,
ptrs_src, ptrs_dst, ptrs_src, ptrs_dst,
ne12, ne13, ne12, ne13,
ne23, ne23,
nb02, nb03, nb02, nb03,
nb12, nb13, nb12, nb13,
dst->nb[2], dst->nb[3], nbd2, nbd3,
r2, r3); r2, r3);
CUDA_CHECK(cudaGetLastError()); CUDA_CHECK(cudaGetLastError());
CUBLAS_CHECK( CUBLAS_CHECK(
cublasGemmBatchedEx(g_cublas_handles[g_main_device], CUBLAS_OP_T, CUBLAS_OP_N, cublasGemmBatchedEx(g_cublas_handles[g_main_device], CUBLAS_OP_T, CUBLAS_OP_N,
ne01, ne11, ne10, ne01, ne11, ne10,
&alpha_f16, (const void **) (ptrs_src + 0*ne23), CUDA_R_16F, nb01/sizeof(half), alpha, (const void **) (ptrs_src + 0*ne23), CUDA_R_16F, nb01/sizeof(half),
(const void **) (ptrs_src + 1*ne23), CUDA_R_16F, nb11/sizeof(float), (const void **) (ptrs_src + 1*ne23), CUDA_R_16F, nb11/sizeof(float),
&beta_f16, ( void **) (ptrs_dst + 0*ne23), CUDA_R_16F, ne01, beta, ( void **) (ptrs_dst + 0*ne23), cu_data_type, ne01,
ne23, ne23,
CUBLAS_COMPUTE_16F, cu_compute_type,
CUBLAS_GEMM_DEFAULT_TENSOR_OP)); CUBLAS_GEMM_DEFAULT_TENSOR_OP));
if (ptrs_src_s != 0) { if (ptrs_src_s != 0) {
@ -8466,11 +8508,14 @@ static void ggml_cuda_mul_mat_mat_batched_cublas(const ggml_tensor * src0, const
} }
#endif #endif
if (dst->op_params[0] == GGML_PREC_DEFAULT) {
const to_fp32_cuda_t to_fp32_cuda = ggml_get_to_fp32_cuda(GGML_TYPE_F16); const to_fp32_cuda_t to_fp32_cuda = ggml_get_to_fp32_cuda(GGML_TYPE_F16);
to_fp32_cuda(dst_f16, dst_ddf, ne, main_stream); to_fp32_cuda(dst_f16, dst_ddf, ne, main_stream);
ggml_cuda_pool_free(src1_as_f16, src1_as);
ggml_cuda_pool_free(dst_f16, dst_as); ggml_cuda_pool_free(dst_f16, dst_as);
}
ggml_cuda_pool_free(src1_as_f16, src1_as);
} }
static void ggml_cuda_mul_mat(const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) { static void ggml_cuda_mul_mat(const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {

13
ggml-metal.metal
View file

@ -1702,8 +1702,9 @@ kernel void kernel_rope(
dst_data[1] = x0*sin_theta + x1*cos_theta; dst_data[1] = x0*sin_theta + x1*cos_theta;
} }
} else { } else {
for (int64_t ib = 0; ib < ne0/n_dims; ++ib) { for (int64_t ic = 2*tiitg; ic < ne0; ic += 2*tptg.x) {
for (int64_t ic = 2*tiitg; ic < n_dims; ic += 2*tptg.x) { if (ic < n_dims) {
const int64_t ib = 0;
// simplified from `(ib * n_dims + ic) * inv_ndims` // simplified from `(ib * n_dims + ic) * inv_ndims`
const float cur_rot = inv_ndims*ic - ib; const float cur_rot = inv_ndims*ic - ib;
@ -1722,6 +1723,14 @@ kernel void kernel_rope(
dst_data[0] = x0*cos_theta - x1*sin_theta; dst_data[0] = x0*cos_theta - x1*sin_theta;
dst_data[n_dims/2] = x0*sin_theta + x1*cos_theta; dst_data[n_dims/2] = x0*sin_theta + x1*cos_theta;
} else {
const int64_t i0 = ic;
device const T * const src = (device T *)((device char *) src0 + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00);
device T * dst_data = (device T *)((device char *) dst + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
dst_data[0] = src[0];
dst_data[1] = src[1];
} }
} }
} }

46
ggml.c
View file

@ -4086,6 +4086,14 @@ struct ggml_tensor * ggml_mul_mat(
return result; return result;
} }
void ggml_mul_mat_set_prec(
struct ggml_tensor * a,
enum ggml_prec prec) {
const int32_t prec_i32 = (int32_t) prec;
ggml_set_op_params_i32(a, 0, prec_i32);
}
// ggml_mul_mat_id // ggml_mul_mat_id
struct ggml_tensor * ggml_mul_mat_id( struct ggml_tensor * ggml_mul_mat_id(
@ -9156,6 +9164,8 @@ static void ggml_compute_forward_norm_f32(
float eps; float eps;
memcpy(&eps, dst->op_params, sizeof(float)); memcpy(&eps, dst->op_params, sizeof(float));
GGML_ASSERT(eps > 0.0f);
// TODO: optimize // TODO: optimize
for (int64_t i03 = 0; i03 < ne03; i03++) { for (int64_t i03 = 0; i03 < ne03; i03++) {
for (int64_t i02 = 0; i02 < ne02; i02++) { for (int64_t i02 = 0; i02 < ne02; i02++) {
@ -9225,6 +9235,8 @@ static void ggml_compute_forward_rms_norm_f32(
float eps; float eps;
memcpy(&eps, dst->op_params, sizeof(float)); memcpy(&eps, dst->op_params, sizeof(float));
GGML_ASSERT(eps > 0.0f);
// TODO: optimize // TODO: optimize
for (int64_t i03 = 0; i03 < ne03; i03++) { for (int64_t i03 = 0; i03 < ne03; i03++) {
for (int64_t i02 = 0; i02 < ne02; i02++) { for (int64_t i02 = 0; i02 < ne02; i02++) {
@ -11550,10 +11562,13 @@ static void ggml_compute_forward_rope_f32(
} }
} else { } else {
// TODO: this might be wrong for ne0 != n_dims - need double check // TODO: this might be wrong for ne0 != n_dims - need double check
// ref: https://github.com/huggingface/transformers/blob/main/src/transformers/models/gpt_neox/modeling_gpt_neox.py#LL251C1-L294C28 // it seems we have to rope just the first n_dims elements and do nothing with the rest
// ref: https://github.com/ml-explore/mlx/blob/dc2edc762c797e3b8de50b1dad4dc0a131691033/benchmarks/python/llama_jax_bench.py#L11-L26
theta_base *= freq_scale; theta_base *= freq_scale;
for (int64_t ib = 0; ib < ne0/n_dims; ++ib) { for (int64_t ic = 0; ic < ne0; ic += 2) {
for (int64_t ic = 0; ic < n_dims; ic += 2) { if (ic < n_dims) {
const int64_t ib = 0;
// simplified from `(ib * n_dims + ic) * inv_ndims` // simplified from `(ib * n_dims + ic) * inv_ndims`
float cur_rot = inv_ndims * ic - ib; float cur_rot = inv_ndims * ic - ib;
@ -11576,6 +11591,14 @@ static void ggml_compute_forward_rope_f32(
dst_data[0] = x0*cos_theta - x1*sin_theta; dst_data[0] = x0*cos_theta - x1*sin_theta;
dst_data[n_dims/2] = x0*sin_theta + x1*cos_theta; dst_data[n_dims/2] = x0*sin_theta + x1*cos_theta;
} else {
const int64_t i0 = ic;
const float * const src = (float *)((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00);
float * dst_data = (float *)((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
dst_data[0] = src[0];
dst_data[1] = src[1];
} }
} }
} }
@ -11703,10 +11726,13 @@ static void ggml_compute_forward_rope_f16(
} }
} else { } else {
// TODO: this might be wrong for ne0 != n_dims - need double check // TODO: this might be wrong for ne0 != n_dims - need double check
// ref: https://github.com/huggingface/transformers/blob/main/src/transformers/models/gpt_neox/modeling_gpt_neox.py#LL251C1-L294C28 // it seems we have to rope just the first n_dims elements and do nothing with the rest
// ref: https://github.com/ml-explore/mlx/blob/dc2edc762c797e3b8de50b1dad4dc0a131691033/benchmarks/python/llama_jax_bench.py#L11-L26
theta_base *= freq_scale; theta_base *= freq_scale;
for (int64_t ib = 0; ib < ne0/n_dims; ++ib) { for (int64_t ic = 0; ic < ne0; ic += 2) {
for (int64_t ic = 0; ic < n_dims; ic += 2) { if (ic < n_dims) {
const int64_t ib = 0;
// simplified from `(ib * n_dims + ic) * inv_ndims` // simplified from `(ib * n_dims + ic) * inv_ndims`
float cur_rot = inv_ndims * ic - ib; float cur_rot = inv_ndims * ic - ib;
@ -11729,6 +11755,14 @@ static void ggml_compute_forward_rope_f16(
dst_data[0] = GGML_FP32_TO_FP16(x0*cos_theta - x1*sin_theta); dst_data[0] = GGML_FP32_TO_FP16(x0*cos_theta - x1*sin_theta);
dst_data[n_dims/2] = GGML_FP32_TO_FP16(x0*sin_theta + x1*cos_theta); dst_data[n_dims/2] = GGML_FP32_TO_FP16(x0*sin_theta + x1*cos_theta);
} else {
const int64_t i0 = ic;
const ggml_fp16_t * const src = (ggml_fp16_t *)((char *) src0->data + i3*nb03 + i2*nb02 + i1*nb01 + i0*nb00);
ggml_fp16_t * dst_data = (ggml_fp16_t *)((char *) dst->data + i3*nb3 + i2*nb2 + i1*nb1 + i0*nb0);
dst_data[0] = src[0];
dst_data[1] = src[1];
} }
} }
} }

12
ggml.h
View file

@ -343,6 +343,12 @@ extern "C" {
GGML_TYPE_COUNT, GGML_TYPE_COUNT,
}; };
// precision
enum ggml_prec {
GGML_PREC_DEFAULT,
GGML_PREC_F32,
};
enum ggml_backend_type { enum ggml_backend_type {
GGML_BACKEND_CPU = 0, GGML_BACKEND_CPU = 0,
GGML_BACKEND_GPU = 10, GGML_BACKEND_GPU = 10,
@ -1057,6 +1063,12 @@ extern "C" {
struct ggml_tensor * a, struct ggml_tensor * a,
struct ggml_tensor * b); struct ggml_tensor * b);
// change the precision of a matrix multiplication
// set to GGML_PREC_F32 for higher precision (useful for phi-2)
GGML_API void ggml_mul_mat_set_prec(
struct ggml_tensor * a,
enum ggml_prec prec);
// indirect matrix multiplication // indirect matrix multiplication
// ggml_mul_mat_id(ctx, as, ids, id, b) ~= ggml_mul_mat(as[ids[id]], b) // ggml_mul_mat_id(ctx, as, ids, id, b) ~= ggml_mul_mat(as[ids[id]], b)
GGML_API struct ggml_tensor * ggml_mul_mat_id( GGML_API struct ggml_tensor * ggml_mul_mat_id(

13
gguf-py/gguf/constants.py
View file

@ -95,6 +95,7 @@ class MODEL_ARCH(IntEnum):
BLOOM = auto() BLOOM = auto()
STABLELM = auto() STABLELM = auto()
QWEN = auto() QWEN = auto()
PHI2 = auto()
class MODEL_TENSOR(IntEnum): class MODEL_TENSOR(IntEnum):
@ -140,6 +141,7 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
MODEL_ARCH.BLOOM: "bloom", MODEL_ARCH.BLOOM: "bloom",
MODEL_ARCH.STABLELM: "stablelm", MODEL_ARCH.STABLELM: "stablelm",
MODEL_ARCH.QWEN: "qwen", MODEL_ARCH.QWEN: "qwen",
MODEL_ARCH.PHI2: "phi2",
} }
TENSOR_NAMES: dict[MODEL_TENSOR, str] = { TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
@ -350,6 +352,17 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
MODEL_ARCH.GPT2: [ MODEL_ARCH.GPT2: [
# TODO # TODO
], ],
MODEL_ARCH.PHI2: [
MODEL_TENSOR.TOKEN_EMBD,
MODEL_TENSOR.OUTPUT_NORM,
MODEL_TENSOR.OUTPUT,
MODEL_TENSOR.ATTN_NORM,
MODEL_TENSOR.ATTN_QKV,
MODEL_TENSOR.ATTN_OUT,
MODEL_TENSOR.FFN_NORM,
MODEL_TENSOR.FFN_DOWN,
MODEL_TENSOR.FFN_UP,
]
# TODO # TODO
} }

8
gguf-py/gguf/tensor_mapping.py
View file

@ -17,6 +17,7 @@ class TensorNameMap:
"tok_embeddings", # llama-pth "tok_embeddings", # llama-pth
"embeddings.word_embeddings", # bert "embeddings.word_embeddings", # bert
"language_model.embedding.word_embeddings", # persimmon "language_model.embedding.word_embeddings", # persimmon
"transformer.embd.wte", # phi2
), ),
# Token type embeddings # Token type embeddings
@ -41,6 +42,7 @@ class TensorNameMap:
"lm_head", # gpt2 mpt falcon llama-hf baichuan qwen "lm_head", # gpt2 mpt falcon llama-hf baichuan qwen
"output", # llama-pth bloom "output", # llama-pth bloom
"word_embeddings_for_head", # persimmon "word_embeddings_for_head", # persimmon
"lm_head.linear", # phi2
), ),
# Output norm # Output norm
@ -53,6 +55,7 @@ class TensorNameMap:
"transformer.norm_f", # mpt "transformer.norm_f", # mpt
"ln_f", # refact bloom qwen "ln_f", # refact bloom qwen
"language_model.encoder.final_layernorm", # persimmon "language_model.encoder.final_layernorm", # persimmon
"lm_head.ln", # phi2
), ),
# Rope frequencies # Rope frequencies
@ -75,6 +78,7 @@ class TensorNameMap:
"encoder.layer.{bid}.attention.output.LayerNorm", # bert "encoder.layer.{bid}.attention.output.LayerNorm", # bert
"language_model.encoder.layers.{bid}.input_layernorm", # persimmon "language_model.encoder.layers.{bid}.input_layernorm", # persimmon
"model.layers.{bid}.ln1", # yi "model.layers.{bid}.ln1", # yi
"transformer.h.{bid}.ln", # phi2
), ),
# Attention norm 2 # Attention norm 2
@ -90,6 +94,7 @@ class TensorNameMap:
"transformer.h.{bid}.self_attention.query_key_value", # falcon "transformer.h.{bid}.self_attention.query_key_value", # falcon
"h.{bid}.self_attention.query_key_value", # bloom "h.{bid}.self_attention.query_key_value", # bloom
"language_model.encoder.layers.{bid}.self_attention.query_key_value", # persimmon "language_model.encoder.layers.{bid}.self_attention.query_key_value", # persimmon
"transformer.h.{bid}.mixer.Wqkv", # phi2
), ),
# Attention query # Attention query
@ -128,6 +133,7 @@ class TensorNameMap:
"encoder.layer.{bid}.attention.output.dense", # bert "encoder.layer.{bid}.attention.output.dense", # bert
"transformer.h.{bid}.attn.out_proj", # gpt-j "transformer.h.{bid}.attn.out_proj", # gpt-j
"language_model.encoder.layers.{bid}.self_attention.dense", # persimmon "language_model.encoder.layers.{bid}.self_attention.dense", # persimmon
"transformer.h.{bid}.mixer.out_proj", # phi2
), ),
# Rotary embeddings # Rotary embeddings
@ -167,6 +173,7 @@ class TensorNameMap:
"transformer.h.{bid}.mlp.fc_in", # gpt-j "transformer.h.{bid}.mlp.fc_in", # gpt-j
"language_model.encoder.layers.{bid}.mlp.dense_h_to_4h", # persimmon "language_model.encoder.layers.{bid}.mlp.dense_h_to_4h", # persimmon
"transformer.h.{bid}.mlp.w1", # qwen "transformer.h.{bid}.mlp.w1", # qwen
"transformer.h.{bid}.mlp.fc1", # phi2
), ),
MODEL_TENSOR.FFN_UP_EXP: ( MODEL_TENSOR.FFN_UP_EXP: (
@ -198,6 +205,7 @@ class TensorNameMap:
"encoder.layer.{bid}.output.dense", # bert "encoder.layer.{bid}.output.dense", # bert
"transformer.h.{bid}.mlp.fc_out", # gpt-j "transformer.h.{bid}.mlp.fc_out", # gpt-j
"language_model.encoder.layers.{bid}.mlp.dense_4h_to_h", # persimmon "language_model.encoder.layers.{bid}.mlp.dense_4h_to_h", # persimmon
"transformer.h.{bid}.mlp.fc2", # phi2
), ),
MODEL_TENSOR.FFN_DOWN_EXP: ( MODEL_TENSOR.FFN_DOWN_EXP: (

287
llama.cpp
View file

@ -195,6 +195,7 @@ enum llm_arch {
LLM_ARCH_BLOOM, LLM_ARCH_BLOOM,
LLM_ARCH_STABLELM, LLM_ARCH_STABLELM,
LLM_ARCH_QWEN, LLM_ARCH_QWEN,
LLM_ARCH_PHI2,
LLM_ARCH_UNKNOWN, LLM_ARCH_UNKNOWN,
}; };
@ -212,6 +213,7 @@ static std::map<llm_arch, std::string> LLM_ARCH_NAMES = {
{ LLM_ARCH_BLOOM, "bloom" }, { LLM_ARCH_BLOOM, "bloom" },
{ LLM_ARCH_STABLELM, "stablelm" }, { LLM_ARCH_STABLELM, "stablelm" },
{ LLM_ARCH_QWEN, "qwen" }, { LLM_ARCH_QWEN, "qwen" },
{ LLM_ARCH_PHI2, "phi2" },
}; };
enum llm_kv { enum llm_kv {
@ -550,6 +552,19 @@ static std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NAMES =
{ LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" }, { LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" },
}, },
}, },
{
LLM_ARCH_PHI2,
{
{ LLM_TENSOR_TOKEN_EMBD, "token_embd" },
{ LLM_TENSOR_OUTPUT_NORM, "output_norm" },
{ LLM_TENSOR_OUTPUT, "output" },
{ LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" },
{ LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" },
{ LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" },
{ LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" },
{ LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" },
},
},
{ {
LLM_ARCH_UNKNOWN, LLM_ARCH_UNKNOWN,
@ -1359,6 +1374,7 @@ struct llama_model {
struct ggml_tensor * output_norm; struct ggml_tensor * output_norm;
struct ggml_tensor * output_norm_b; struct ggml_tensor * output_norm_b;
struct ggml_tensor * output; struct ggml_tensor * output;
struct ggml_tensor * output_b;
std::vector<llama_layer> layers; std::vector<llama_layer> layers;
@ -1873,7 +1889,7 @@ namespace GGUFMeta {
target = override->bool_value; target = override->bool_value;
return true; return true;
} }
return true; return false;
} }
template<typename OT> template<typename OT>
@ -2573,6 +2589,15 @@ static void llm_load_hparams(
default: model.type = e_model::MODEL_UNKNOWN; default: model.type = e_model::MODEL_UNKNOWN;
} }
} break; } break;
case LLM_ARCH_PHI2:
{
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
switch (hparams.n_layer) {
case 32: model.type = e_model::MODEL_3B; break;
default: model.type = e_model::MODEL_UNKNOWN;
}
} break;
default: (void)0; default: (void)0;
} }
@ -3419,7 +3444,57 @@ static void llm_load_tensors(
layer.ffn_up = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, backend_split); layer.ffn_up = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, backend_split);
} }
} break; } break;
case LLM_ARCH_PHI2:
{
model.tok_embd = ml.create_tensor(ctx, tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, GGML_BACKEND_CPU);
// output
{
ggml_backend_type backend_norm;
ggml_backend_type backend_output;
if (n_gpu_layers > int(n_layer)) {
backend_norm = llama_backend_offload;
backend_output = llama_backend_offload;
} else {
backend_norm = GGML_BACKEND_CPU;
backend_output = GGML_BACKEND_CPU;
}
model.output_norm = ml.create_tensor(ctx, tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, backend_norm);
model.output_norm_b = ml.create_tensor(ctx, tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, backend_norm);
model.output = ml.create_tensor(ctx, tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, backend_output);
model.output_b = ml.create_tensor(ctx, tn(LLM_TENSOR_OUTPUT, "bias"), {n_vocab}, backend_output);
}
const uint32_t n_ff = hparams.n_ff;
const int i_gpu_start = n_layer - n_gpu_layers;
model.layers.resize(n_layer);
for (uint32_t i = 0; i < n_layer; ++i) {
const ggml_backend_type backend = int(i) < i_gpu_start ? GGML_BACKEND_CPU : llama_backend_offload; // NOLINT
const ggml_backend_type backend_split = int(i) < i_gpu_start ? GGML_BACKEND_CPU : llama_backend_offload_split; // NOLINT
auto & layer = model.layers[i];
layer.attn_norm = ml.create_tensor(ctx, tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, backend);
layer.attn_norm_b = ml.create_tensor(ctx, tn(LLM_TENSOR_ATTN_NORM, "bias", i), {n_embd}, backend);
layer.wqkv = ml.create_tensor(ctx, tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, backend_split);
layer.bqkv = ml.create_tensor(ctx, tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, backend);
layer.wo = ml.create_tensor(ctx, tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, backend_split);
layer.bo = ml.create_tensor(ctx, tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, backend);
layer.ffn_down = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_DOWN, "weight", i), {n_ff, n_embd}, backend_split);
layer.ffn_down_b = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, backend);
layer.ffn_up = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, backend_split);
layer.ffn_up_b = ml.create_tensor(ctx, tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, backend);
}
} break;
default: default:
throw std::runtime_error("unknown architecture"); throw std::runtime_error("unknown architecture");
} }
@ -3824,6 +3899,7 @@ static struct ggml_tensor * llm_build_ffn(
// if max_alibi_bias > 0 then apply ALiBi // if max_alibi_bias > 0 then apply ALiBi
static struct ggml_tensor * llm_build_kqv( static struct ggml_tensor * llm_build_kqv(
struct ggml_context * ctx, struct ggml_context * ctx,
const llama_model & model,
const llama_hparams & hparams, const llama_hparams & hparams,
const llama_kv_cache & kv, const llama_kv_cache & kv,
struct ggml_tensor * wo, struct ggml_tensor * wo,
@ -3835,6 +3911,7 @@ static struct ggml_tensor * llm_build_kqv(
int32_t n_tokens, int32_t n_tokens,
int32_t n_kv, int32_t n_kv,
float max_alibi_bias, float max_alibi_bias,
float scale,
const llm_build_cb & cb, const llm_build_cb & cb,
int il) { int il) {
const int64_t n_embd = hparams.n_embd; const int64_t n_embd = hparams.n_embd;
@ -3857,6 +3934,12 @@ static struct ggml_tensor * llm_build_kqv(
struct ggml_tensor * kq = ggml_mul_mat(ctx, k, q); struct ggml_tensor * kq = ggml_mul_mat(ctx, k, q);
cb(kq, "kq", il); cb(kq, "kq", il);
if (model.arch == LLM_ARCH_PHI2) {
// for this arch, we need to perform the KQ multiplication with F32 precision, otherwise we get NaNs
// ref: https://github.com/ggerganov/llama.cpp/pull/4490#issuecomment-1859055847
ggml_mul_mat_set_prec(kq, GGML_PREC_F32);
}
if (max_alibi_bias > 0.0f) { if (max_alibi_bias > 0.0f) {
// temporary branch until we figure out how to handle ggml_alibi through ggml_add // temporary branch until we figure out how to handle ggml_alibi through ggml_add
kq = ggml_scale(ctx, kq, kq_scale); kq = ggml_scale(ctx, kq, kq_scale);
@ -3876,7 +3959,7 @@ static struct ggml_tensor * llm_build_kqv(
kq = ggml_soft_max(ctx, kq); kq = ggml_soft_max(ctx, kq);
cb(kq, "kq_soft_max", il); cb(kq, "kq_soft_max", il);
} else { } else {
kq = ggml_soft_max_ext(ctx, kq, kq_mask, 1.0f/sqrtf(float(n_embd_head))); kq = ggml_soft_max_ext(ctx, kq, kq_mask, scale);
cb(kq, "kq_soft_max_ext", il); cb(kq, "kq_soft_max_ext", il);
} }
@ -4083,9 +4166,9 @@ struct llm_build_context {
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il); llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);
cur = llm_build_kqv(ctx0, hparams, kv_self, cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, model.layers[il].bo, model.layers[il].wo, model.layers[il].bo,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, cb, il); Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, 1.0f/sqrtf(float(n_embd_head)), cb, il);
cb(cur, "kqv_out", il); cb(cur, "kqv_out", il);
} }
@ -4266,9 +4349,9 @@ struct llm_build_context {
// apply ALiBi for 13B model // apply ALiBi for 13B model
const float max_alibi_bias = model.type == MODEL_13B ? 8.0f : -1.0f; const float max_alibi_bias = model.type == MODEL_13B ? 8.0f : -1.0f;
cur = llm_build_kqv(ctx0, hparams, kv_self, cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, NULL, model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, max_alibi_bias, cb, il); Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, max_alibi_bias, 1.0f/sqrtf(float(n_embd_head)), cb, il);
cb(cur, "kqv_out", il); cb(cur, "kqv_out", il);
} }
@ -4390,9 +4473,9 @@ struct llm_build_context {
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il); llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);
cur = llm_build_kqv(ctx0, hparams, kv_self, cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, NULL, model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, cb, il); Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, 1.0f/sqrtf(float(n_embd_head)), cb, il);
cb(cur, "kqv_out", il); cb(cur, "kqv_out", il);
} }
@ -4490,9 +4573,9 @@ struct llm_build_context {
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il); llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);
cur = llm_build_kqv(ctx0, hparams, kv_self, cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, model.layers[il].bo, model.layers[il].wo, model.layers[il].bo,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, cb, il); Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, 1.0f/sqrtf(float(n_embd_head)), cb, il);
cb(cur, "kqv_out", il); cb(cur, "kqv_out", il);
} }
@ -4699,9 +4782,9 @@ struct llm_build_context {
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il); llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);
// TODO: not tested, could be broken // TODO: not tested, could be broken
cur = llm_build_kqv(ctx0, hparams, kv_self, cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, model.layers[il].bo, model.layers[il].wo, model.layers[il].bo,
Q, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, cb, il); Q, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, 1.0f/sqrtf(float(n_embd_head)), cb, il);
cb(cur, "kqv_out", il); cb(cur, "kqv_out", il);
} }
@ -4790,9 +4873,9 @@ struct llm_build_context {
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il); llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);
cur = llm_build_kqv(ctx0, hparams, kv_self, cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, NULL, model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, 8.0f, cb, il); Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, 8.0f, 1.0f/sqrtf(float(n_embd_head)), cb, il);
cb(cur, "kqv_out", il); cb(cur, "kqv_out", il);
} }
@ -4887,9 +4970,9 @@ struct llm_build_context {
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il); llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);
cur = llm_build_kqv(ctx0, hparams, kv_self, cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, model.layers[il].bo, model.layers[il].wo, model.layers[il].bo,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, 8.0f, cb, il); Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, 8.0f, 1.0f/sqrtf(float(n_embd_head)), cb, il);
cb(cur, "kqv_out", il); cb(cur, "kqv_out", il);
} }
@ -4981,9 +5064,9 @@ struct llm_build_context {
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il); llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);
cur = llm_build_kqv(ctx0, hparams, kv_self, cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, NULL, model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, hparams.f_max_alibi_bias, cb, il); Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, hparams.f_max_alibi_bias, 1.0f/sqrtf(float(n_embd_head)), cb, il);
cb(cur, "kqv_out", il); cb(cur, "kqv_out", il);
} }
@ -5094,9 +5177,9 @@ struct llm_build_context {
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il); llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);
cur = llm_build_kqv(ctx0, hparams, kv_self, cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, NULL, model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, cb, il); Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, 1.0f/sqrtf(float(n_embd_head)), cb, il);
cb(cur, "kqv_out", il); cb(cur, "kqv_out", il);
} }
@ -5153,15 +5236,15 @@ struct llm_build_context {
cb(inpL, "inp_embd", -1); cb(inpL, "inp_embd", -1);
// inp_pos - contains the positions // inp_pos - contains the positions
struct ggml_tensor * inp_pos= ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_tokens); struct ggml_tensor * inp_pos = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_tokens);
cb(inp_pos, "inp_pos", -1); cb(inp_pos, "inp_pos", -1);
// KQ_scale // KQ_scale
struct ggml_tensor * KQ_scale= ggml_new_tensor_1d(ctx0, GGML_TYPE_F32, 1); struct ggml_tensor * KQ_scale = ggml_new_tensor_1d(ctx0, GGML_TYPE_F32, 1);
cb(KQ_scale, "KQ_scale", -1); cb(KQ_scale, "KQ_scale", -1);
// KQ_mask (mask for 1 head, it will be broadcasted to all heads) // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
struct ggml_tensor * KQ_mask= ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_kv, n_tokens, 1); struct ggml_tensor * KQ_mask = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_kv, n_tokens, 1);
cb(KQ_mask, "KQ_mask", -1); cb(KQ_mask, "KQ_mask", -1);
// shift the entire K-cache if needed // shift the entire K-cache if needed
@ -5211,9 +5294,9 @@ struct llm_build_context {
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il); llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);
cur = llm_build_kqv(ctx0, hparams, kv_self, cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, NULL, model.layers[il].wo, NULL,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, cb, il); Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, 1.0f/sqrtf(float(n_embd_head)), cb, il);
cb(cur, "kqv_out", il); cb(cur, "kqv_out", il);
} }
@ -5255,6 +5338,122 @@ struct llm_build_context {
ggml_build_forward_expand(gf, cur); ggml_build_forward_expand(gf, cur);
return gf;
}
struct ggml_cgraph * build_phi2() {
struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false);
struct ggml_tensor * cur;
struct ggml_tensor * attn_norm_output;
struct ggml_tensor * ffn_output;
struct ggml_tensor * inpL;
inpL = llm_build_inp_embd(ctx0, hparams, batch, model.tok_embd, cb);
cb(inpL, "inp_embd", -1);
// inp_pos - contains the positions
struct ggml_tensor * inp_pos = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_tokens);
cb(inp_pos, "inp_pos", -1);
// Q_scale
struct ggml_tensor * Q_scale = ggml_new_tensor_1d(ctx0, GGML_TYPE_F32, 1);
cb(Q_scale, "Q_scale", -1);
// KQ_scale
struct ggml_tensor * KQ_scale = ggml_new_tensor_1d(ctx0, GGML_TYPE_F32, 1);
cb(KQ_scale, "KQ_scale", -1);
// KQ_mask (mask for 1 head, it will be broadcasted to all heads)
struct ggml_tensor * KQ_mask = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, n_kv, n_tokens, 1);
cb(KQ_mask, "KQ_mask", -1);
// shift the entire K-cache if needed
if (do_rope_shift) {
llm_build_k_shift(ctx0, hparams, cparams, kv_self, gf, LLM_ROPE_NEOX, n_ctx, n_embd_head, freq_base, freq_scale, cb);
}
for (int il = 0; il < n_layer; ++il) {
attn_norm_output = llm_build_norm(ctx0, inpL, hparams,
model.layers[il].attn_norm,
model.layers[il].attn_norm_b,
LLM_NORM, cb, il);
cb(attn_norm_output, "attn_norm", il);
// self-attention
{
cur = ggml_mul_mat(ctx0, model.layers[il].wqkv, attn_norm_output);
cb(cur, "wqkv", il);
cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
cb(cur, "bqkv", il);
struct ggml_tensor * Qcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd, n_tokens, cur->nb[1], 0*sizeof(float)*(n_embd)));
struct ggml_tensor * Kcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd)));
struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_2d(ctx0, cur, n_embd_gqa, n_tokens, cur->nb[1], 1*sizeof(float)*(n_embd + n_embd_gqa)));
cb(Qcur, "Qcur", il);
cb(Kcur, "Kcur", il);
cb(Vcur, "Vcur", il);
Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head, n_tokens);
Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
Qcur = ggml_rope_custom(
ctx0, Qcur, inp_pos, hparams.n_rot, 2, 0, n_orig_ctx,
freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
);
cb(Qcur, "Qcur", il);
Qcur = ggml_scale(ctx0, Qcur, Q_scale);
cb(Qcur, "Qcur", il);
Kcur = ggml_rope_custom(
ctx0, Kcur, inp_pos, hparams.n_rot, 2, 0, n_orig_ctx,
freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
);
cb(Kcur, "Kcur", il);
llm_build_kv_store(ctx0, hparams, kv_self, gf, Kcur, Vcur, n_ctx, n_tokens, kv_head, cb, il);
cur = llm_build_kqv(ctx0, model, hparams, kv_self,
model.layers[il].wo, model.layers[il].bo,
Qcur, KQ_scale, KQ_mask, n_ctx, n_tokens, n_kv, -1.0f, 1.0f, cb, il);
cb(cur, "kqv_out", il);
}
// FF
{
ffn_output = llm_build_ffn(ctx0, attn_norm_output,
model.layers[il].ffn_up, model.layers[il].ffn_up_b,
NULL, NULL,
model.layers[il].ffn_down, model.layers[il].ffn_down_b,
LLM_FFN_GELU, LLM_FFN_SEQ, cb, il);
cb(ffn_output, "ffn_out", il);
}
cur = ggml_add(ctx0, cur, ffn_output);
cb(cur, "l_out", il);
cur = ggml_add(ctx0, cur, inpL);
cb(cur, "l_out", il);
inpL = cur;
}
cur = llm_build_norm(ctx0, inpL, hparams,
model.output_norm,
model.output_norm_b,
LLM_NORM, cb, -1);
cb(cur, "result_norm", -1);
cur = ggml_mul_mat(ctx0, model.output, cur);
cb(cur, "result_output_no_bias", -1);
cur = ggml_add(ctx0, cur, model.output_b);
cb(cur, "result_output", -1);
ggml_build_forward_expand(gf, cur);
return gf; return gf;
} }
}; };
@ -5270,7 +5469,7 @@ enum llm_offload_func_e {
OFFLOAD_FUNC_FRC, // force offload OFFLOAD_FUNC_FRC, // force offload
OFFLOAD_FUNC_KQV, OFFLOAD_FUNC_KQV,
OFFLOAD_FUNC_NR, OFFLOAD_FUNC_NR,
OFFLOAD_FUNC_EMB, OFFLOAD_FUNC_EMB, // embeddings
OFFLOAD_FUNC_OUT, OFFLOAD_FUNC_OUT,
}; };
@ -5355,6 +5554,7 @@ static const std::unordered_map<const char *, llm_offload_func_e> k_offload_map
{ "pos_embd", OFFLOAD_FUNC_NR }, { "pos_embd", OFFLOAD_FUNC_NR },
{ "inp_pos", OFFLOAD_FUNC_FRC }, // this is often used for KQ ops (e.g. rope) { "inp_pos", OFFLOAD_FUNC_FRC }, // this is often used for KQ ops (e.g. rope)
{ "Q_scale", OFFLOAD_FUNC_FRC },
{ "KQ_scale", OFFLOAD_FUNC_FRC }, { "KQ_scale", OFFLOAD_FUNC_FRC },
{ "KQ_mask", OFFLOAD_FUNC_FRC }, { "KQ_mask", OFFLOAD_FUNC_FRC },
{ "K_shift", OFFLOAD_FUNC_FRC }, { "K_shift", OFFLOAD_FUNC_FRC },
@ -5439,6 +5639,7 @@ static const std::unordered_map<const char *, llm_offload_func_e> k_offload_map
{ "l_out", OFFLOAD_FUNC }, { "l_out", OFFLOAD_FUNC },
{ "result_norm", OFFLOAD_FUNC_EMB }, { "result_norm", OFFLOAD_FUNC_EMB },
{ "result_output_no_bias", OFFLOAD_FUNC_EMB },
{ "result_output", OFFLOAD_FUNC_OUT }, { "result_output", OFFLOAD_FUNC_OUT },
}; };
@ -5456,6 +5657,7 @@ static struct ggml_cgraph * llama_build_graph(
bool alloc_inp_tokens = false; bool alloc_inp_tokens = false;
bool alloc_inp_embd = false; bool alloc_inp_embd = false;
bool alloc_inp_pos = false; bool alloc_inp_pos = false;
bool alloc_inp_Q_scale = false;
bool alloc_inp_KQ_scale = false; bool alloc_inp_KQ_scale = false;
bool alloc_inp_KQ_mask = false; bool alloc_inp_KQ_mask = false;
bool alloc_inp_K_shift = false; bool alloc_inp_K_shift = false;
@ -5523,7 +5725,7 @@ static struct ggml_cgraph * llama_build_graph(
alloc_inp_pos = true; alloc_inp_pos = true;
} }
if (!alloc_inp_KQ_scale && strcmp(name, "KQ_scale") == 0) { if (!alloc_inp_Q_scale && strcmp(name, "Q_scale") == 0) {
ggml_allocr_alloc(lctx.alloc, cur); ggml_allocr_alloc(lctx.alloc, cur);
if (!ggml_allocr_is_measure(lctx.alloc)) { if (!ggml_allocr_is_measure(lctx.alloc)) {
@ -5531,6 +5733,23 @@ static struct ggml_cgraph * llama_build_graph(
ggml_set_f32(cur, 1.0f/sqrtf(float(n_embd_head))); ggml_set_f32(cur, 1.0f/sqrtf(float(n_embd_head)));
} }
alloc_inp_Q_scale = true;
}
if (!alloc_inp_KQ_scale && strcmp(name, "KQ_scale") == 0) {
ggml_allocr_alloc(lctx.alloc, cur);
if (!ggml_allocr_is_measure(lctx.alloc)) {
const int64_t n_embd_head = model.hparams.n_embd_head();
if (model.arch == LLM_ARCH_PHI2) {
// with phi2, we scale the Q to avoid precision issues
// ref: https://github.com/ml-explore/mlx-examples/blob/08e862336ade809bc37d1035f94b359e7d1a5152/phi2/phi2.py#L64-L66
ggml_set_f32(cur, 1.0f);
} else {
ggml_set_f32(cur, 1.0f/sqrtf(float(n_embd_head)));
}
}
alloc_inp_KQ_scale = true; alloc_inp_KQ_scale = true;
} }
@ -5755,6 +5974,10 @@ static struct ggml_cgraph * llama_build_graph(
{ {
result = llm.build_qwen(); result = llm.build_qwen();
} break; } break;
case LLM_ARCH_PHI2:
{
result = llm.build_phi2();
} break;
default: default:
GGML_ASSERT(false); GGML_ASSERT(false);
} }
@ -5888,12 +6111,16 @@ static int llama_decode_internal(
ggml_allocr_alloc_graph(lctx.alloc, gf); ggml_allocr_alloc_graph(lctx.alloc, gf);
// the output is always the last tensor in the graph
struct ggml_tensor * res = gf->nodes[gf->n_nodes - 1]; struct ggml_tensor * res = gf->nodes[gf->n_nodes - 1];
struct ggml_tensor * embeddings = gf->nodes[gf->n_nodes - 2];
GGML_ASSERT(strcmp(res->name, "result_output") == 0); GGML_ASSERT(strcmp(res->name, "result_output") == 0);
GGML_ASSERT(strcmp(embeddings->name, "result_norm") == 0);
// the embeddings could be the second to last tensor, or the third to last tensor
struct ggml_tensor * embeddings = gf->nodes[gf->n_nodes - 2];
if (strcmp(embeddings->name, "result_norm") != 0) {
embeddings = gf->nodes[gf->n_nodes - 3];
GGML_ASSERT(strcmp(embeddings->name, "result_norm") == 0);
}
#ifdef GGML_USE_CUBLAS #ifdef GGML_USE_CUBLAS
char * buf_alloc_base = (char *)ggml_backend_buffer_get_base(lctx.buf_alloc); char * buf_alloc_base = (char *)ggml_backend_buffer_get_base(lctx.buf_alloc);

1
tests/test-backend-ops.cpp
View file

@ -1555,6 +1555,7 @@ static bool test_backend(ggml_backend_t backend, test_mode mode, const char * op
test_cases.emplace_back(new test_rope(type, { 64, 8, 10, 1}, 64, 2, 512)); // neox (falcon 40B) test_cases.emplace_back(new test_rope(type, { 64, 8, 10, 1}, 64, 2, 512)); // neox (falcon 40B)
test_cases.emplace_back(new test_rope(type, { 64, 128, 10, 1}, 64, 2, 512)); // neox (falcon 40B) test_cases.emplace_back(new test_rope(type, { 64, 128, 10, 1}, 64, 2, 512)); // neox (falcon 40B)
test_cases.emplace_back(new test_rope(type, { 80, 32, 10, 1}, 20, 2, 512)); // neox (stablelm) test_cases.emplace_back(new test_rope(type, { 80, 32, 10, 1}, 20, 2, 512)); // neox (stablelm)
test_cases.emplace_back(new test_rope(type, { 80, 32, 10, 1}, 32, 2, 512)); // neox (phi-2)
} }
test_cases.emplace_back(new test_alibi()); test_cases.emplace_back(new test_alibi());