diff --git a/olmocr/train/grpo_train.py b/olmocr/train/grpo_train.py index b0c4822..ec7106a 100644 --- a/olmocr/train/grpo_train.py +++ b/olmocr/train/grpo_train.py @@ -280,7 +280,7 @@ def main(): parser.add_argument( "--gradient_accumulation_steps", type=int, - default=4, + default=8, help="Gradient accumulation steps" ) parser.add_argument(