From ce85c789f044dbad70876b162137eb0bb9541077 Mon Sep 17 00:00:00 2001 From: Yu Shi Jie Date: Thu, 4 Sep 2025 22:31:47 -0400 Subject: [PATCH] initial commit --- trl/trainer/grpo_trainer.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/trl/trainer/grpo_trainer.py b/trl/trainer/grpo_trainer.py index 9af99e73a76..32f5efb154d 100644 --- a/trl/trainer/grpo_trainer.py +++ b/trl/trainer/grpo_trainer.py @@ -418,6 +418,8 @@ def __init__( compute_loss_func="non-None value to disable scaling", ) + self.current_gradient_accumulation_steps = args.gradient_accumulation_steps + # Reference model self.beta = args.beta if self.beta == 0.0: