From abc16ad45f4187a97b8e8cdeed3883ec4d2ebac8 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Mon, 17 Feb 2025 19:53:49 -0800 Subject: [PATCH] Update rl_replacements.py --- unsloth/models/rl_replacements.py | 1 + 1 file changed, 1 insertion(+) diff --git a/unsloth/models/rl_replacements.py b/unsloth/models/rl_replacements.py index 3d97b90df9..17215bafb8 100644 --- a/unsloth/models/rl_replacements.py +++ b/unsloth/models/rl_replacements.py @@ -201,6 +201,7 @@ RL_FUNCTIONS["grpo_trainer"].append(grpo_trainer__get_per_token_logps) grpo_compute_loss = RL_REPLACEMENTS["grpo_compute_loss"] grpo_accumulated_loss = RL_REPLACEMENTS["grpo_accumulated_loss"] RL_PRE_ITEMS["grpo_trainer"].append(inspect.getsource(grpo_compute_loss)) +RL_PRE_ITEMS["grpo_accumulated_loss"].append(inspect.getsource(grpo_accumulated_loss)) # Edit _get_per_token_logps to handle mixed precision def grpo_trainer_compute_loss(function_name, function):