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

llama_model_loader: be sure the model mappings has enough capacity before allocating backend buffer

Browse source
This commit is contained in:
Pierrick HYMBERT 2024-03-22 07:03:14 +01:00
parent a9e88c6e57
commit b19af3643f
1 changed files with 2 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

2
llama.cpp
View file

@ -5135,9 +5135,11 @@ static bool llm_load_tensors(
ml.done_getting_tensors(); ml.done_getting_tensors();
ml.init_mappings(true, &model.mlock_mmaps); ml.init_mappings(true, &model.mlock_mmaps);
model.mappings.reserve(ml.mappings.size());
// create the backend buffers // create the backend buffers
std::vector<std::pair<ggml_context *, std::unordered_map<uint32_t, ggml_backend_buffer_t>>> ctx_bufs; std::vector<std::pair<ggml_context *, std::unordered_map<uint32_t, ggml_backend_buffer_t>>> ctx_bufs;
ctx_bufs.reserve(ctx_map.size());
// Ensure we have enough capacity for the maximum backend buffer we will potentially create // Ensure we have enough capacity for the maximum backend buffer we will potentially create
size_t n_max_backend_buffer = ctx_map.size() * ml.files.size(); size_t n_max_backend_buffer = ctx_map.size() * ml.files.size();