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

support report custom args

Browse Source 
Former-commit-id: d41254c40a1c5cacf9377096adb27efa9bdb79ea
This commit is contained in:
hiyouga
2024-12-19 14:57:09 +00:00
parent adff887659
commit a897d46049
20 changed files with 164 additions and 124 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/llamafactory/train/ppo/trainer.py
View File

@@ -40,7 +40,7 @@ from typing_extensions import override
from ...extras import logging
from ...extras.misc import AverageMeter, count_parameters, get_current_device, get_logits_processor
from ..callbacks import FixValueHeadModelCallback, SaveProcessorCallback
from ..trainer_utils import create_custom_optimizer, create_custom_scheduler, get_swanlab_callback
from ..trainer_utils import create_custom_optimizer, create_custom_scheduler
from .ppo_utils import dump_layernorm, get_rewards_from_server, replace_model, restore_layernorm
@@ -186,9 +186,6 @@ class CustomPPOTrainer(PPOTrainer, Trainer):
self.accelerator.clip_grad_norm_ = MethodType(clip_grad_norm_old_version, self.accelerator)
self.add_callback(BAdamCallback)
if finetuning_args.use_swanlab:
self.add_callback(get_swanlab_callback(finetuning_args))
def ppo_train(self, resume_from_checkpoint: Optional[str] = None) -> None:
r"""
Implements training loop for the PPO stage, like _inner_training_loop() in Huggingface's Trainer.