vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-10 09:44:38 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

add mmap prefetch and unloading

Browse source
This commit is contained in:
Lizonghang 2024-10-25 16:33:56 +04:00
parent ba5117581e
commit c97ea10617
7 changed files with 161 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

1
include/llama.h
View file

@ -318,6 +318,7 @@ extern "C" {
uint32_t n_world; // world size
uint32_t rank; // my rank
uint32_t n_layer_window; // number of layers to process in each compute
bool unload; // whether to unload layer weights after use
char * master_ip; // ip address of the master node
char * next_node_ip; // ip address of the next node
uint32_t n_ctx; // text context, 0 = from model