tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

order matters

Browse Source
This commit is contained in:
Wing Lian
2025-02-03 00:47:48 -05:00
parent 300ffc2cb6
commit c810599c66
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/axolotl/core/trainer_builder.py
View File

@@ -1019,7 +1019,7 @@ class HFRLTrainerBuilder(TrainerBuilderBase):
training_args_kwargs.update(DPOConfig.set_training_args_kwargs(self.cfg))
training_args = training_args_cls( # pylint: disable=unexpected-keyword-arg
output_dir=self.cfg.output_dir,
self.cfg.output_dir,
per_device_train_batch_size=self.cfg.micro_batch_size,
max_steps=self.cfg.max_steps or total_num_steps,
gradient_accumulation_steps=self.cfg.gradient_accumulation_steps,

2
src/axolotl/core/trainers/grpo/args.py
View File

@@ -9,7 +9,7 @@ from axolotl.core.training_args import AxolotlTrainingMixins
@dataclass
class AxolotlGRPOConfig(GRPOConfig, AxolotlTrainingMixins):
class AxolotlGRPOConfig(AxolotlTrainingMixins, GRPOConfig):
"""
Axolotl GRPO Config for GRPO training
"""