mmarch/text-generation-webui
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Add --pin-weight parameter for FlexGen

Browse source
This commit is contained in:
oobabooga 2023-03-04 01:04:02 -03:00
parent 05e703b4a4
commit ed8b35efd2
2 changed files with 12 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
modules/models.py
View file

@ -58,7 +58,7 @@ def load_model(model_name):
shared.args.percent[0], shared.args.percent[1],
shared.args.percent[2], shared.args.percent[3],
shared.args.percent[4], shared.args.percent[5],
overlap=True, sep_layer=True, pin_weight=True,
overlap=True, sep_layer=True, pin_weight=shared.args.pin_weight,
cpu_cache_compute=False, attn_sparsity=1.0,
compress_weight=shared.args.compress_weight,
comp_weight_config=CompressionConfig(