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

Add --no-op-offload to improve -ot pp perf in MoE models like llama4 400B (#13386)

Browse source
This commit is contained in:
David Huang 2025-05-11 20:18:39 +08:00 • committed by GitHub
parent 3eac209319
commit 7f323a589f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
11 changed files with 57 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

1
common/common.h
View file

@ -332,6 +332,7 @@ struct common_params {
bool no_kv_offload = false; // disable KV offloading
bool warmup = true; // warmup run
bool check_tensors = false; // validate tensor data
bool no_op_offload = false; // globally disable offload host tensor operations to device
bool single_turn = false; // single turn chat conversation