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

revert unrelated changed in cuda cmake

Browse source 
remove useless nommq
fix typo of GGML_USE_CLBLAS_SYCL
This commit is contained in:
Meng, Hengyu 2024-01-24 12:18:30 +00:00
parent 67de350e79
commit fb15de38ef
2 changed files with 12 additions and 16 deletions
Download patch file Download diff file Expand all files Collapse all files

6
CMakeLists.txt
View file

@ -345,10 +345,16 @@ if (LLAMA_CUBLAS)
add_compile_definitions(GGML_CUDA_PEER_MAX_BATCH_SIZE=${LLAMA_CUDA_PEER_MAX_BATCH_SIZE}) add_compile_definitions(GGML_CUDA_PEER_MAX_BATCH_SIZE=${LLAMA_CUDA_PEER_MAX_BATCH_SIZE})
if (LLAMA_STATIC) if (LLAMA_STATIC)
if (WIN32)
# As of 12.3.1 CUDA Tookit for Windows does not offer a static cublas library
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas CUDA::cublasLt)
else ()
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas_static CUDA::cublasLt_static) set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart_static CUDA::cublas_static CUDA::cublasLt_static)
endif()
else() else()
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart CUDA::cublas CUDA::cublasLt) set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cudart CUDA::cublas CUDA::cublasLt)
endif() endif()
set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} CUDA::cuda_driver)
if (NOT DEFINED CMAKE_CUDA_ARCHITECTURES) if (NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
# 52 == lowest CUDA 12 standard # 52 == lowest CUDA 12 standard

20
common/common.cpp
View file

@ -603,9 +603,9 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
break; break;
} }
params.main_gpu = std::stoi(argv[i]); params.main_gpu = std::stoi(argv[i]);
#ifndef GGML_USE_CLBLAS_SYCL #ifndef GGML_USE_CUBLAS_SYCL
fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS/SYCL. Setting the main GPU has no effect.\n"); fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS/SYCL. Setting the main GPU has no effect.\n");
#endif // GGML_USE_CLBLAS_SYCL #endif // GGML_USE_CUBLAS_SYCL
} else if (arg == "--split-mode" || arg == "-sm") { } else if (arg == "--split-mode" || arg == "-sm") {
if (++i >= argc) { if (++i >= argc) {
invalid_param = true; invalid_param = true;
@ -622,9 +622,9 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
invalid_param = true; invalid_param = true;
break; break;
} }
#ifndef GGML_USE_CLBLAS_SYCL #ifndef GGML_USE_CUBLAS_SYCL
fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS/SYCL. Setting the split mode has no effect.\n"); fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS/SYCL. Setting the split mode has no effect.\n");
#endif // GGML_USE_CLBLAS_SYCL #endif // GGML_USE_CUBLAS_SYCL
} else if (arg == "--tensor-split" || arg == "-ts") { } else if (arg == "--tensor-split" || arg == "-ts") {
if (++i >= argc) { if (++i >= argc) {
@ -651,7 +651,7 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
} }
#else #else
fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS/SYCL. Setting a tensor split has no effect.\n"); fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS/SYCL. Setting a tensor split has no effect.\n");
#endif // GGML_USE_CLBLAS_SYCL #endif // GGML_USE_CUBLAS_SYCL
} else if (arg == "--no-mul-mat-q" || arg == "-nommq") { } else if (arg == "--no-mul-mat-q" || arg == "-nommq") {
#if defined(GGML_USE_CUBLAS) || defined(GGML_USE_SYCL) #if defined(GGML_USE_CUBLAS) || defined(GGML_USE_SYCL)
params.mul_mat_q = false; params.mul_mat_q = false;
@ -1020,16 +1020,6 @@ void gpt_print_usage(int /*argc*/, char ** argv, const gpt_params & params) {
printf(" -mg i, --main-gpu i the GPU to use for the model (with split-mode = none),\n"); printf(" -mg i, --main-gpu i the GPU to use for the model (with split-mode = none),\n");
printf(" or for intermediate results and KV (with split-mode = row) (default: %d)\n", params.main_gpu); printf(" or for intermediate results and KV (with split-mode = row) (default: %d)\n", params.main_gpu);
#endif // LLAMA_SUPPORTS_GPU_OFFLOAD #endif // LLAMA_SUPPORTS_GPU_OFFLOAD
#ifdef GGML_USE_CLBLAS
printf(" -nommq, --no-mul-mat-q\n");
printf(" use " GGML_CUBLAS_NAME " instead of custom mul_mat_q " GGML_CUDA_NAME " kernels.\n");
printf(" Not recommended since this is both slower and uses more VRAM.\n");
#endif // GGML_USE_CLBLAS
#ifdef GGML_USE_SYCL
printf(" -nommq, --no-mul-mat-q\n");
printf(" use " GGML_SYCL_NAME " instead of custom mul_mat_q " GGML_SYCL_NAME " kernels.\n");
printf(" Not recommended since this is both slower and uses more VRAM.\n");
#endif // GGML_USE_SYCL
printf(" --verbose-prompt print a verbose prompt before generation (default: %s)\n", params.verbose_prompt ? "true" : "false"); printf(" --verbose-prompt print a verbose prompt before generation (default: %s)\n", params.verbose_prompt ? "true" : "false");
printf(" --no-display-prompt don't print prompt at generation (default: %s)\n", !params.display_prompt ? "true" : "false"); printf(" --no-display-prompt don't print prompt at generation (default: %s)\n", !params.display_prompt ? "true" : "false");
printf(" -gan N, --grp-attn-n N\n"); printf(" -gan N, --grp-attn-n N\n");