ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Load all MoE experts during warmup (#11571)

Browse source 
* llama : introduce llama_set_warmup() API call that controls warmup mode; use all MoE experts during warmup

* common : use new API to enable warmup mode during model warmup

---------

Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
This commit is contained in:
fairydreaming 2025-03-14 13:47:05 +01:00 • committed by GitHub
parent add2a3aa5a
commit 8fcb563613
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 22 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
include/llama.h
View file

@ -945,6 +945,10 @@ extern "C" {
// If set to true, the model will only attend to the past tokens
LLAMA_API void llama_set_causal_attn(struct llama_context * ctx, bool causal_attn);
// Set whether the model is in warmup mode or not
// If true, all model tensors are activated during llama_decode() to load and cache their weights.
LLAMA_API void llama_set_warmup(struct llama_context * ctx, bool warmup);
// Set abort callback
LLAMA_API void llama_set_abort_callback(struct llama_context * ctx, ggml_abort_callback abort_callback, void * abort_callback_data);