ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-02 18:10:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

Change --n-cpu-moe to not keep expert biases ion CPU

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-10-19 15:21:43 +03:00
parent 747f411da5
commit d7319055de
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
common/common.cpp
View File

@@ -1164,7 +1164,7 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
return true;
}
for (int32_t l = 0; l < n_layers; ++l) {
std::string pattern = "blk\\." + std::to_string(l) + "\\.(ffn_(up|down|gate)_exps)";
std::string pattern = "blk\\." + std::to_string(l) + "\\.(ffn_(up|down|gate)_exps\\.weight)";
params.tensor_buft_overrides.push_back({strdup(pattern.c_str()), ggml_backend_cpu_buffer_type()});
}
return true;