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

Fix crash when using split mode none and setting a main GPU

Browse source
This commit is contained in:
0cc4m 2024-06-02 09:50:02 +02:00
parent 6e0e0beb56
commit fe3f6958bd
2 changed files with 7 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

12
common/common.cpp
View file

@ -1002,9 +1002,9 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
return true; return true;
} }
params.main_gpu = std::stoi(argv[i]); params.main_gpu = std::stoi(argv[i]);
#ifndef GGML_USE_CUDA_SYCL #ifndef GGML_USE_CUDA_SYCL_VULKAN
fprintf(stderr, "warning: llama.cpp was compiled without CUDA/SYCL. Setting the main GPU has no effect.\n"); fprintf(stderr, "warning: llama.cpp was compiled without CUDA/SYCL/Vulkan. Setting the main GPU has no effect.\n");
#endif // GGML_USE_CUDA_SYCL #endif // GGML_USE_CUDA_SYCL_VULKAN
return true; return true;
} }
if (arg == "--split-mode" || arg == "-sm") { if (arg == "--split-mode" || arg == "-sm") {
@ -1030,9 +1030,9 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
invalid_param = true; invalid_param = true;
return true; return true;
} }
#ifndef GGML_USE_CUDA_SYCL #ifndef GGML_USE_CUDA_SYCL_VULKAN
fprintf(stderr, "warning: llama.cpp was compiled without CUDA/SYCL. Setting the split mode has no effect.\n"); fprintf(stderr, "warning: llama.cpp was compiled without CUDA/SYCL/Vulkan. Setting the split mode has no effect.\n");
#endif // GGML_USE_CUDA_SYCL #endif // GGML_USE_CUDA_SYCL_VULKAN
return true; return true;
} }
if (arg == "--tensor-split" || arg == "-ts") { if (arg == "--tensor-split" || arg == "-ts") {

2
llama.cpp
View file

@ -16353,7 +16353,7 @@ struct llama_context * llama_new_context_with_model(
return nullptr; return nullptr;
} }
if (model->split_mode == LLAMA_SPLIT_MODE_NONE) { if (model->split_mode == LLAMA_SPLIT_MODE_NONE) {
ggml_backend_t backend = ggml_backend_vk_init(0); ggml_backend_t backend = ggml_backend_vk_init(model->main_gpu);
if (backend == nullptr) { if (backend == nullptr) {
LLAMA_LOG_ERROR("%s: failed to initialize Vulkan backend\n", __func__); LLAMA_LOG_ERROR("%s: failed to initialize Vulkan backend\n", __func__);
llama_free(ctx); llama_free(ctx);