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

llama : fix compatibility with old 2 expert models

Browse source
This commit is contained in:
slaren 2024-04-18 03:52:39 +02:00
parent 3b8f1ec4b1
commit ce80217fa9
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
llama.cpp
View file

@ -4592,7 +4592,7 @@ static bool llm_load_tensors(
size_t ctx_size = ggml_tensor_overhead()*(ml.n_tensors + 1); // +1 for models where tok_embd is duplicated as output
// for moe merged tensors
ctx_size += ggml_tensor_overhead()*hparams.n_expert*n_layer;
ctx_size += ggml_tensor_overhead()*n_layer*3;
std::map<ggml_backend_buffer_type_t, ggml_context *> ctx_map;
for (auto & it : buft_layer_count) {