llama test
This commit is contained in:
@@ -13,7 +13,7 @@ liger-kernel==0.5.2
|
||||
packaging==23.2
|
||||
|
||||
peft==0.14.0
|
||||
transformers==4.47.1
|
||||
transformers==4.48.0
|
||||
tokenizers>=0.21.0
|
||||
accelerate==1.2.1
|
||||
datasets==3.2.0
|
||||
|
||||
@@ -387,12 +387,13 @@ class ModelLoader:
|
||||
self.patch_attention()
|
||||
|
||||
if self.cfg.model_config_type == "llama":
|
||||
from axolotl.monkeypatch.trainer_grad_accum import ( # patch_flash_attention_forward,
|
||||
from axolotl.monkeypatch.trainer_grad_accum import (
|
||||
patch_flash_attention_forward,
|
||||
patch_forward_for_ga,
|
||||
patch_training_step_for_ga,
|
||||
)
|
||||
|
||||
# patch_flash_attention_forward()
|
||||
patch_flash_attention_forward()
|
||||
patch_forward_for_ga()
|
||||
patch_training_step_for_ga()
|
||||
|
||||
|
||||
Reference in New Issue
Block a user