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

llama : abort ctx if cuda backend init fails

Browse source
This commit is contained in:
slaren 2024-01-08 14:00:34 +01:00
parent 5a62db30c3
commit 4813e17548
1 changed files with 5 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

6
llama.cpp
View file

@ -9199,10 +9199,12 @@ struct llama_context * llama_new_context_with_model(
#elif defined(GGML_USE_CUBLAS) #elif defined(GGML_USE_CUBLAS)
if (model->n_gpu_layers > 0) { if (model->n_gpu_layers > 0) {
// with split_mode LLAMA_SPLIT_NONE or LLAMA_SPLIT_ROW, only the main GPU backend is used // with split_mode LLAMA_SPLIT_NONE or LLAMA_SPLIT_ROW, only the main GPU backend is used
if (model->split_mode == LLAMA_SPLIT_ROW || model->split_mode == LLAMA_SPLIT_NONE) { if (model->split_mode == LLAMA_SPLIT_NONE || model->split_mode == LLAMA_SPLIT_ROW) {
ggml_backend_t backend = ggml_backend_cuda_init(model->main_gpu); ggml_backend_t backend = ggml_backend_cuda_init(model->main_gpu);
if (backend == nullptr) { if (backend == nullptr) {
LLAMA_LOG_ERROR("%s: failed to initialize CUDA%d backend\n", __func__, model->main_gpu); LLAMA_LOG_ERROR("%s: failed to initialize CUDA%d backend\n", __func__, model->main_gpu);
llama_free(ctx);
return nullptr;
} }
ctx->backends.push_back(backend); ctx->backends.push_back(backend);
} else { } else {
@ -9211,6 +9213,8 @@ struct llama_context * llama_new_context_with_model(
ggml_backend_t backend = ggml_backend_cuda_init(device); ggml_backend_t backend = ggml_backend_cuda_init(device);
if (backend == nullptr) { if (backend == nullptr) {
LLAMA_LOG_ERROR("%s: failed to initialize CUDA%d backend\n", __func__, device); LLAMA_LOG_ERROR("%s: failed to initialize CUDA%d backend\n", __func__, device);
llama_free(ctx);
return nullptr;
} }
ctx->backends.push_back(backend); ctx->backends.push_back(backend);
} }