Add --no_xformers and --no_sdpa flags for ExllamaV2

This commit is contained in:
oobabooga 2024-07-11 15:47:37 -07:00
parent 512b311137
commit e436d69e2b
6 changed files with 15 additions and 1 deletions

View file

@ -84,6 +84,8 @@ loaders_and_params = OrderedDict({
'max_seq_len',
'cfg_cache',
'no_flash_attn',
'no_xformers',
'no_sdpa',
'num_experts_per_token',
'cache_8bit',
'cache_4bit',
@ -97,6 +99,8 @@ loaders_and_params = OrderedDict({
'gpu_split',
'max_seq_len',
'no_flash_attn',
'no_xformers',
'no_sdpa',
'num_experts_per_token',
'cache_8bit',
'cache_4bit',