ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix ppo trainer

Browse Source 
Former-commit-id: ca5b5823b03822ef899405d233a82396be997f44
This commit is contained in:
hiyouga
2023-12-28 18:09:28 +08:00
parent bf381563ff
commit 53d7c5109f
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llmtuner/train/ppo/trainer.py
View File

@@ -203,7 +203,7 @@ class CustomPPOTrainer(PPOTrainer, Trainer):
r""" r"""
Generates model's responses given queries. Generates model's responses given queries.
""" """
if self.finetuning_args.upcast_layernorm: if self.model_args.upcast_layernorm:
layernorm_params = dump_layernorm(self.model) layernorm_params = dump_layernorm(self.model)
if batch["input_ids"].size(0) == 1: # handle llama2 ppo with gradient accumulation > 1 if batch["input_ids"].size(0) == 1: # handle llama2 ppo with gradient accumulation > 1
@@ -218,7 +218,7 @@ class CustomPPOTrainer(PPOTrainer, Trainer):
**batch **batch
) )
if self.finetuning_args.upcast_layernorm: if self.model_args.upcast_layernorm:
restore_layernorm(self.model, layernorm_params) restore_layernorm(self.model, layernorm_params)
query = batch["input_ids"].detach().cpu() query = batch["input_ids"].detach().cpu()