vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-10 12:54:34 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

rename arg --keep-inp-out-in-metal to --keep-out-in-metal

Browse source
This commit is contained in:
Lizonghang 2025-01-23 23:17:06 +04:00
parent 5fcf020cfb
commit 1c0087e919
5 changed files with 14 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

6
common/common.cpp
View file

@ -1162,7 +1162,7 @@ static void assign_device(
if (dev.gpu_support.cuda || dev.gpu_support.metal) {
float reserved_mem = 0.1f; // reserved shared memory to avoid potential OOM, set to 100 MiB by default
vec_z_gpu[m] = (double)((dev.gpu_props.memory_free - reserved_mem) * GIGABYTE - c_gpu[m]) / (double)(n_layer * b_prime);
if (dev.gpu_support.metal && m == 0 && cparams.keep_inp_out_in_metal) {
if (dev.gpu_support.metal && m == 0 && cparams.keep_out_in_metal) {
vec_z_gpu[m] -= (double)(bi + bo) / (double)(n_layer * b_prime);
}
dev_gpu[m] = 1;
@ -1624,7 +1624,7 @@ struct llama_model_params llama_model_params_from_gpt_params(const gpt_params &
mparams.use_mmap = params.use_mmap;
mparams.use_mlock = params.use_mlock;
mparams.check_tensors = params.check_tensors;
mparams.keep_inp_out_in_metal = params.keep_inp_out_in_metal;
mparams.keep_out_in_metal = params.keep_out_in_metal;
std::copy(std::begin(params.n_layer_window), std::end(params.n_layer_window), mparams.n_layer_window);
if (params.kv_overrides.empty()) {
mparams.kv_overrides = NULL;
@ -1671,7 +1671,7 @@ struct llama_context_params llama_context_params_from_gpt_params(const gpt_param
cparams.n_world = params.n_world;
cparams.rank = params.rank;
cparams.unload = params.unload;
cparams.keep_inp_out_in_metal = params.keep_inp_out_in_metal;
cparams.keep_out_in_metal = params.keep_out_in_metal;
cparams.n_gpu_layers = params.n_gpu_layers;
std::copy(std::begin(params.n_layer_window), std::end(params.n_layer_window), cparams.n_layer_window);