tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

simplify logic (#1856)

Browse Source
This commit is contained in:
Wing Lian
2024-08-23 20:23:08 -04:00
committed by GitHub
parent 77a4b9cda2
commit 22f4eafa55
1 changed files with 6 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
src/axolotl/utils/models.py
View File

@@ -589,15 +589,8 @@ def load_model(
# sample packing uses custom FA2 patch # sample packing uses custom FA2 patch
if cfg.flash_attention: if cfg.flash_attention:
if not cfg.sample_packing: if not cfg.sample_packing and cfg.s2_attention:
if cfg.s2_attention:
pass pass
# most other models support flash attention, we can define exceptions as they come up
model_kwargs["attn_implementation"] = "flash_attention_2"
model_config._attn_implementation = ( # pylint: disable=protected-access
"flash_attention_2"
)
else:
model_kwargs["attn_implementation"] = "flash_attention_2" model_kwargs["attn_implementation"] = "flash_attention_2"
model_config._attn_implementation = ( # pylint: disable=protected-access model_config._attn_implementation = ( # pylint: disable=protected-access
"flash_attention_2" "flash_attention_2"