Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 14:20:31 +01:00
Code Issues Packages Projects Releases Wiki Activity

backup

Browse Source
This commit is contained in:
Meng, Hengyu 2024-05-31 16:51:56 +08:00
parent 1f80e0e428
commit d32a8f6142
5 changed files with 209 additions and 263 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

448
ggml-sycl.cpp
View File

File diff suppressed because it is too large Load Diff

9
ggml-sycl.h
View File

@ -13,6 +13,8 @@
extern "C" { extern "C" {
#endif #endif
#define GGML_SYCL_MAX_DEVICES 48
// backend API // backend API
GGML_API ggml_backend_t ggml_backend_sycl_init(int device); GGML_API ggml_backend_t ggml_backend_sycl_init(int device);
@ -30,13 +32,6 @@ GGML_API GGML_CALL void ggml_sycl_get_gpu_list(int *id_list, int max_len);
GGML_API GGML_CALL void ggml_sycl_get_device_description(int device, char *description, size_t description_size); GGML_API GGML_CALL void ggml_sycl_get_device_description(int device, char *description, size_t description_size);
GGML_API GGML_CALL int ggml_backend_sycl_get_device_count(); GGML_API GGML_CALL int ggml_backend_sycl_get_device_count();
GGML_API GGML_CALL void ggml_backend_sycl_get_device_memory(int device, size_t *free, size_t *total); GGML_API GGML_CALL void ggml_backend_sycl_get_device_memory(int device, size_t *free, size_t *total);
GGML_API GGML_CALL int ggml_backend_sycl_get_device_index(int device_id);
// TODO: these are temporary
// ref: https://github.com/ggerganov/llama.cpp/pull/6022#issuecomment-1992615670
GGML_API GGML_CALL int ggml_backend_sycl_get_device_id(int device_index);
GGML_API GGML_CALL void ggml_backend_sycl_set_single_device_mode(int main_gpu_id);
GGML_API GGML_CALL void ggml_backend_sycl_set_mul_device_mode();
// SYCL doesn't support registering host memory, keep here for reference // SYCL doesn't support registering host memory, keep here for reference
// GGML_API GGML_CALL bool ggml_backend_sycl_register_host_buffer(void * buffer, size_t size); // GGML_API GGML_CALL bool ggml_backend_sycl_register_host_buffer(void * buffer, size_t size);

1
ggml-sycl/common.hpp
View File

@ -136,6 +136,7 @@ typedef sycl::float2 dfloat2;
static const int8_t kvalues_iq4nl[16]={-127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113}; static const int8_t kvalues_iq4nl[16]={-127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113};
static int g_all_sycl_device_count = -1;
static bool g_ggml_backend_sycl_buffer_type_initialized = false; static bool g_ggml_backend_sycl_buffer_type_initialized = false;
static ggml_sycl_backend_gpu_mode g_ggml_sycl_backend_gpu_mode = static ggml_sycl_backend_gpu_mode g_ggml_sycl_backend_gpu_mode =

1
ggml-sycl/presets.hpp
View File

@ -13,7 +13,6 @@
#ifndef GGML_SYCL_PRESETS_HPP #ifndef GGML_SYCL_PRESETS_HPP
#define GGML_SYCL_PRESETS_HPP #define GGML_SYCL_PRESETS_HPP
#define GGML_SYCL_MAX_DEVICES 48
#define GGML_SYCL_MAX_STREAMS 8 #define GGML_SYCL_MAX_STREAMS 8
#define GGML_SYCL_MAX_BUFFERS 256 #define GGML_SYCL_MAX_BUFFERS 256
#define GGML_SYCL_NAME "SYCL" #define GGML_SYCL_NAME "SYCL"

13
llama.cpp
View File

@ -6518,16 +6518,6 @@ static int llama_model_load(const std::string & fname, llama_model & model, llam
} }
#endif #endif
#ifdef GGML_USE_SYCL
if (params.split_mode == LLAMA_SPLIT_MODE_NONE) {
ggml_backend_sycl_set_single_device_mode(params.main_gpu);
//SYCL use device index (0, 1, 2) directly, uer input device id, then convert to device index.
params.main_gpu = ggml_backend_sycl_get_device_index(params.main_gpu);
} else {
ggml_backend_sycl_set_mul_device_mode();
}
#endif
if (!llm_load_tensors( if (!llm_load_tensors(
ml, model, params.n_gpu_layers, params.split_mode, params.main_gpu, params.tensor_split, params.use_mlock, ml, model, params.n_gpu_layers, params.split_mode, params.main_gpu, params.tensor_split, params.use_mlock,
params.progress_callback, params.progress_callback_user_data params.progress_callback, params.progress_callback_user_data
@ -16376,8 +16366,7 @@ struct llama_context * llama_new_context_with_model(
if (model->split_mode == LLAMA_SPLIT_MODE_NONE || model->split_mode == LLAMA_SPLIT_MODE_ROW) { if (model->split_mode == LLAMA_SPLIT_MODE_NONE || model->split_mode == LLAMA_SPLIT_MODE_ROW) {
ggml_backend_t backend = ggml_backend_sycl_init(model->main_gpu); ggml_backend_t backend = ggml_backend_sycl_init(model->main_gpu);
if (backend == nullptr) { if (backend == nullptr) {
int main_gpu_id = ggml_backend_sycl_get_device_id(model->main_gpu); LLAMA_LOG_ERROR("%s: failed to initialize SYCL%d backend\n", __func__, model->main_gpu);
LLAMA_LOG_ERROR("%s: failed to initialize SYCL%d (index %d) backend\n", __func__, main_gpu_id, model->main_gpu);
llama_free(ctx); llama_free(ctx);
return nullptr; return nullptr;
} }