Set use_fast=True by default, create --no_use_fast flag

This increases tokens/second for HF loaders.
This commit is contained in:
oobabooga 2023-11-16 19:45:05 -08:00
parent b2ce8dc7ee
commit 8b66d83aa9
6 changed files with 20 additions and 19 deletions

View file

@ -52,7 +52,7 @@ def list_model_elements():
'bf16',
'load_in_8bit',
'trust_remote_code',
'use_fast',
'no_use_fast',
'use_flash_attention_2',
'load_in_4bit',
'compute_dtype',