AI/gpt4all
1
0
Fork 0
mirror of https://github.com/nomic-ai/gpt4all.git synced 2024-10-01 01:06:10 -04:00
Code Issues Packages Projects Releases Wiki Activity

Fix VRAM leak when model loading fails (#1901)

Browse Source 
Signed-off-by: Jared Van Bortel <jared@nomic.ai>
This commit is contained in:
Jared Van Bortel 2024-02-01 15:45:45 -05:00 committed by GitHub
parent e1eac00ee0
commit 10e3f7bbf5
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 16 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
gpt4all-backend/llama.cpp-mainline

@ -1 +1 @@
Subproject commit 47aec1bcc09e090f0b8f196dc0a4e43b89507e4a Subproject commit cd1b5a104b9d3e211a50b9f6c261aced3bf09834

16
gpt4all-backend/llamamodel.cpp
View File

@ -150,7 +150,15 @@ size_t LLamaModel::requiredMem(const std::string &modelPath, int n_ctx, int ngl)
bool LLamaModel::loadModel(const std::string &modelPath, int n_ctx, int ngl) bool LLamaModel::loadModel(const std::string &modelPath, int n_ctx, int ngl)
{ {
gpt_params params; // clean up after previous loadModel()
if (d_ptr->model) {
llama_free_model(d_ptr->model);
d_ptr->model = nullptr;
}
if (d_ptr->ctx) {
llama_free(d_ptr->ctx);
d_ptr->ctx = nullptr;
}
if (n_ctx < 8) { if (n_ctx < 8) {
std::cerr << "warning: minimum context size is 8, using minimum size.\n"; std::cerr << "warning: minimum context size is 8, using minimum size.\n";
@ -159,6 +167,8 @@ bool LLamaModel::loadModel(const std::string &modelPath, int n_ctx, int ngl)
// -- load the model -- // -- load the model --
gpt_params params;
d_ptr->model_params = llama_model_default_params(); d_ptr->model_params = llama_model_default_params();
d_ptr->model_params.use_mmap = params.use_mmap; d_ptr->model_params.use_mmap = params.use_mmap;
@ -215,8 +225,10 @@ bool LLamaModel::loadModel(const std::string &modelPath, int n_ctx, int ngl)
d_ptr->ctx = llama_new_context_with_model(d_ptr->model, d_ptr->ctx_params); d_ptr->ctx = llama_new_context_with_model(d_ptr->model, d_ptr->ctx_params);
if (!d_ptr->ctx) { if (!d_ptr->ctx) {
d_ptr->device = -1;
std::cerr << "LLAMA ERROR: failed to init context for model " << modelPath << std::endl; std::cerr << "LLAMA ERROR: failed to init context for model " << modelPath << std::endl;
llama_free_model(d_ptr->model);
d_ptr->model = nullptr;
d_ptr->device = -1;
return false; return false;
} }

1
gpt4all-chat/chatllm.cpp
View File

@ -228,6 +228,7 @@ bool ChatLLM::loadModel(const ModelInfo &modelInfo)
LLModelStore::globalInstance()->releaseModel(m_llModelInfo); // release back into the store LLModelStore::globalInstance()->releaseModel(m_llModelInfo); // release back into the store
m_llModelInfo = LLModelInfo(); m_llModelInfo = LLModelInfo();
emit modelLoadingError(QString("Previous attempt to load model resulted in crash for `%1` most likely due to insufficient memory. You should either remove this model or decrease your system RAM usage by closing other applications.").arg(modelInfo.filename())); emit modelLoadingError(QString("Previous attempt to load model resulted in crash for `%1` most likely due to insufficient memory. You should either remove this model or decrease your system RAM usage by closing other applications.").arg(modelInfo.filename()));
return false;
} }
if (fileInfo.exists()) { if (fileInfo.exists()) {