From 8aa722a140392f4657d378fdea1075a0a84d0dca Mon Sep 17 00:00:00 2001 From: NanoCode012 Date: Fri, 21 Feb 2025 13:55:34 +0700 Subject: [PATCH] fix: ignore max_length for grpo --- src/axolotl/core/trainers/grpo/__init__.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/src/axolotl/core/trainers/grpo/__init__.py b/src/axolotl/core/trainers/grpo/__init__.py index f4685893b..1b240c6b4 100644 --- a/src/axolotl/core/trainers/grpo/__init__.py +++ b/src/axolotl/core/trainers/grpo/__init__.py @@ -131,8 +131,8 @@ class GRPOStrategy: return None @classmethod - def get_blocklist_args_kwargs(cls) -> list[str]: - return ["dataset_num_proc"] + def get_blocklist_args_kwargs(cls): + return ["dataset_num_proc", "max_length"] @classmethod def get_reward_func(cls, reward_func_fqn: str) -> RewardFunc: @@ -167,4 +167,4 @@ class GRPOStrategy: LOG.info( f"Reward function {reward_func_fqn} is a pre-trained model path - if this is unexpected, please check the reward function path." ) - return reward_func + return reward_func_fqn