vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-10 04:54:34 +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
include/llama.h
View file

@ -322,7 +322,7 @@ extern "C" {
uint32_t rank; // my rank
uint32_t n_layer_window[32];// number of layers to process in each compute
uint32_t n_gpu_layers; // number of layers to process on GPU
bool unload; // whether to unload layer weights after use
bool prefetch; // whether to prefetch layer weights
bool keep_out_in_metal; // whether to keep output weights in metal memory
char * master_ip; // ip address of the master node
char * next_node_ip; // ip address of the next node