mirror of
https://github.com/nomic-ai/gpt4all.git
synced 2025-09-06 11:00:48 +00:00
Fallback to CPU more robustly.
This commit is contained in:
Submodule gpt4all-backend/llama.cpp-mainline updated: 7ff671e149...703ef9c125
@@ -168,6 +168,10 @@ bool LLamaModel::loadModel(const std::string &modelPath)
|
||||
|
||||
d_ptr->ctx = llama_init_from_file(modelPath.c_str(), d_ptr->params);
|
||||
if (!d_ptr->ctx) {
|
||||
#ifdef GGML_USE_KOMPUTE
|
||||
// Explicitly free the device so next load it doesn't use it
|
||||
ggml_vk_free_device();
|
||||
#endif
|
||||
std::cerr << "LLAMA ERROR: failed to load model from " << modelPath << std::endl;
|
||||
return false;
|
||||
}
|
||||
@@ -194,7 +198,7 @@ int32_t LLamaModel::threadCount() const {
|
||||
|
||||
LLamaModel::~LLamaModel()
|
||||
{
|
||||
if(d_ptr->ctx) {
|
||||
if (d_ptr->ctx) {
|
||||
llama_free(d_ptr->ctx);
|
||||
}
|
||||
}
|
||||
|
Reference in New Issue
Block a user