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

llama : add option to override tensor buffers

Browse source
This commit is contained in:
slaren 2025-01-24 20:56:09 +01:00
parent 9fbadaef4f
commit f07c2ec505
9 changed files with 87 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

10
common/common.cpp
View file

@ -1083,15 +1083,18 @@ struct llama_model_params common_model_params_to_llama(common_params & params) {
if (!params.devices.empty()) {
mparams.devices = params.devices.data();
}
if (params.n_gpu_layers != -1) {
mparams.n_gpu_layers = params.n_gpu_layers;
}
mparams.main_gpu = params.main_gpu;
mparams.split_mode = params.split_mode;
mparams.tensor_split = params.tensor_split;
mparams.use_mmap = params.use_mmap;
mparams.use_mlock = params.use_mlock;
mparams.check_tensors = params.check_tensors;
if (params.kv_overrides.empty()) {
mparams.kv_overrides = NULL;
} else {
@ -1099,6 +1102,13 @@ struct llama_model_params common_model_params_to_llama(common_params & params) {
mparams.kv_overrides = params.kv_overrides.data();
}
if (params.tensor_buft_overrides.empty()) {
mparams.tensor_buft_overrides = NULL;
} else {
GGML_ASSERT(params.tensor_buft_overrides.back().pattern == nullptr && "Tensor buffer overrides not terminated with empty pattern");
mparams.tensor_buft_overrides = params.tensor_buft_overrides.data();
}
return mparams;
}