vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-10 18:24:34 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

use arg prefetch and remove arg unload

Browse source
This commit is contained in:
Lizonghang 2025-02-12 17:04:41 +04:00
parent 708b1d8c89
commit c84f9d29fe
5 changed files with 15 additions and 19 deletions
Download patch file Download diff file Expand all files Collapse all files

2
common/common.h
View file

@ -147,7 +147,7 @@ struct gpt_params {
uint32_t n_layer_window[32] = {0}; // layer window size on each node
std::string master_ip = "localhost"; // ip address of the master node
std::string next_node_ip = "localhost"; // ip address of my next node
bool unload = false; // unload layer weights after use or not
bool prefetch = false; // prefetch layer weights
bool keep_out_in_metal = true; // whether to keep output weights in metal memory, true by default
int32_t gpu_mem = 999.0; // gpu memory to use, in GiB
int32_t n_predict = -1; // new tokens to predict