423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-10-15 16:18:10 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix ppo trainer

Browse Source 
Former-commit-id: a03b2e5ef0d5d6b1b27753438745385d290cb211
This commit is contained in:
hiyouga 2024-07-10 11:05:45 +08:00
parent 834c4e8ad9
commit 446129ca7a
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/llamafactory/train/ppo/trainer.py
View File

@ -106,7 +106,8 @@ class CustomPPOTrainer(PPOTrainer, Trainer):
DistributedDataParallelKwargs(find_unused_parameters=training_args.ddp_find_unused_parameters) DistributedDataParallelKwargs(find_unused_parameters=training_args.ddp_find_unused_parameters)
] ]
ppo_config.accelerator_kwargs["deepspeed_plugin"] = training_args.deepspeed_plugin ppo_config.accelerator_kwargs["deepspeed_plugin"] = training_args.deepspeed_plugin
if ppo_config.log_with == "tensorboard": # tensorboard raises error about accelerator_kwargs if ppo_config.log_with is not None:
logger.warning("PPOTrainer cannot use external logger when DeepSpeed is enabled.")
ppo_config.log_with = None ppo_config.log_with = None
# Create optimizer and scheduler # Create optimizer and scheduler