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

add args -k and --force

Browse source
This commit is contained in:
Zonghang Li 2025-03-11 20:44:36 +04:00
parent 9cbdf01645
commit bcfdace59b
5 changed files with 27 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

2
include/llama.h
View file

@ -324,7 +324,9 @@ 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
uint32_t n_cycles; // number of cycles to output one token
bool prefetch; // whether to prefetch layer weights
bool force; // force to start prefetching after computation
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