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

minor

Browse source
This commit is contained in:
slaren 2024-01-08 13:42:20 +01:00
parent 444b975edd
commit d41cef9326
3 changed files with 20 additions and 28 deletions
Download patch file Download diff file Expand all files Collapse all files

30
common/common.cpp
View file

@ -529,9 +529,8 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
invalid_param = true; invalid_param = true;
break; break;
} }
#ifdef LLAMA_SUPPORTS_GPU_OFFLOAD
params.n_gpu_layers = std::stoi(argv[i]); params.n_gpu_layers = std::stoi(argv[i]);
#else #ifndef LLAMA_SUPPORTS_GPU_OFFLOAD
fprintf(stderr, "warning: not compiled with GPU offload support, --n-gpu-layers option will be ignored\n"); fprintf(stderr, "warning: not compiled with GPU offload support, --n-gpu-layers option will be ignored\n");
fprintf(stderr, "warning: see main README.md for information on enabling GPU BLAS support\n"); fprintf(stderr, "warning: see main README.md for information on enabling GPU BLAS support\n");
#endif #endif
@ -540,9 +539,8 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
invalid_param = true; invalid_param = true;
break; break;
} }
#ifdef LLAMA_SUPPORTS_GPU_OFFLOAD
params.n_gpu_layers_draft = std::stoi(argv[i]); params.n_gpu_layers_draft = std::stoi(argv[i]);
#else #ifndef LLAMA_SUPPORTS_GPU_OFFLOAD
fprintf(stderr, "warning: not compiled with GPU offload support, --n-gpu-layers-draft option will be ignored\n"); fprintf(stderr, "warning: not compiled with GPU offload support, --n-gpu-layers-draft option will be ignored\n");
fprintf(stderr, "warning: see main README.md for information on enabling GPU BLAS support\n"); fprintf(stderr, "warning: see main README.md for information on enabling GPU BLAS support\n");
#endif #endif
@ -551,11 +549,10 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
invalid_param = true; invalid_param = true;
break; break;
} }
#ifdef GGML_USE_CUBLAS
params.main_gpu = std::stoi(argv[i]); params.main_gpu = std::stoi(argv[i]);
#else #ifndef GGML_USE_CUBLAS
fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS. It is not possible to set a main GPU.\n"); fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS. Setting the main GPU has no effect.\n");
#endif #endif // GGML_USE_CUBLAS
} 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;
@ -580,15 +577,16 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
invalid_param = true; invalid_param = true;
break; break;
} }
#ifdef GGML_USE_CUBLAS
std::string arg_next = argv[i]; std::string arg_next = argv[i];
// split string by , and / // split string by , and /
const std::regex regex{R"([,/]+)"}; const std::regex regex{R"([,/]+)"};
std::sregex_token_iterator it{arg_next.begin(), arg_next.end(), regex, -1}; std::sregex_token_iterator it{arg_next.begin(), arg_next.end(), regex, -1};
std::vector<std::string> split_arg{it, {}}; std::vector<std::string> split_arg{it, {}};
GGML_ASSERT(split_arg.size() <= LLAMA_MAX_DEVICES); if (split_arg.size() >= LLAMA_MAX_DEVICES) {
invalid_param = true;
break;
}
for (size_t i = 0; i < LLAMA_MAX_DEVICES; ++i) { for (size_t i = 0; i < LLAMA_MAX_DEVICES; ++i) {
if (i < split_arg.size()) { if (i < split_arg.size()) {
params.tensor_split[i] = std::stof(split_arg[i]); params.tensor_split[i] = std::stof(split_arg[i]);
@ -596,14 +594,8 @@ bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params) {
params.tensor_split[i] = 0.0f; params.tensor_split[i] = 0.0f;
} }
} }
#else #ifndef GGML_USE_CUBLAS
fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS. It is not possible to set a tensor split.\n"); fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS. Setting a tensor split has no effect.\n");
#endif // GGML_USE_CUBLAS
} else if (arg == "--no-mul-mat-q" || arg == "-nommq") {
#ifdef GGML_USE_CUBLAS
params.mul_mat_q = false;
#else
fprintf(stderr, "warning: llama.cpp was compiled without cuBLAS. Disabling mul_mat_q kernels has no effect.\n");
#endif // GGML_USE_CUBLAS #endif // GGML_USE_CUBLAS
} else if (arg == "--no-mmap") { } else if (arg == "--no-mmap") {
params.use_mmap = false; params.use_mmap = false;

4
llama.h
View file

@ -117,7 +117,7 @@ extern "C" {
enum llama_split_mode { enum llama_split_mode {
LLAMA_SPLIT_NONE = 0, // single GPU LLAMA_SPLIT_NONE = 0, // single GPU
LLAMA_SPLIT_LAYER = 1, // split layers and KV to different GPUs LLAMA_SPLIT_LAYER = 1, // split layers and KV across GPUs
LLAMA_SPLIT_ROW = 2, // split rows across GPUs LLAMA_SPLIT_ROW = 2, // split rows across GPUs
}; };
@ -188,7 +188,7 @@ extern "C" {
// for small tensors and intermediate results (LLAMA_SPLIT_ROW) // for small tensors and intermediate results (LLAMA_SPLIT_ROW)
// ignored for LLAMA_SPLIT_LAYER // ignored for LLAMA_SPLIT_LAYER
int32_t main_gpu; int32_t main_gpu;
// fraction of the model (layers or rows) to offload to each GPU, size: LLAMA_MAX_DEVICES // proportion of the model (layers or rows) to offload to each GPU, size: LLAMA_MAX_DEVICES
const float * tensor_split; const float * tensor_split;
// Called with a progress value between 0.0 and 1.0. Pass NULL to disable. // Called with a progress value between 0.0 and 1.0. Pass NULL to disable.

14
tests/test-backend-ops.cpp
View file

@ -474,17 +474,17 @@ struct test_case {
printf("compare failed "); printf("compare failed ");
} }
if (ud.ok && cmp_ok) {
printf("\033[1;32mOK\033[0m\n");
} else {
printf("\033[1;31mFAIL\033[0m\n");
}
ggml_backend_buffer_free(buf); ggml_backend_buffer_free(buf);
ggml_free(ctx); ggml_free(ctx);
return ud.ok; if (ud.ok && cmp_ok) {
printf("\033[1;32mOK\033[0m\n");
return true;
}
printf("\033[1;31mFAIL\033[0m\n");
return false;
} }
bool eval_perf(ggml_backend_t backend, const char * op_name) { bool eval_perf(ggml_backend_t backend, const char * op_name) {