From 18dfabd8e5ddf4996fba8f78f8a9a4901faa9da0 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Wed, 19 Feb 2025 22:03:47 -0800 Subject: [PATCH] Update rl_replacements.py --- unsloth/models/rl_replacements.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/unsloth/models/rl_replacements.py b/unsloth/models/rl_replacements.py index c8caa1b585..5d6201dd22 100644 --- a/unsloth/models/rl_replacements.py +++ b/unsloth/models/rl_replacements.py @@ -200,8 +200,10 @@ pass RL_FUNCTIONS["grpo_trainer"].append(grpo_trainer__get_per_token_logps) grpo_compute_loss = RL_REPLACEMENTS["grpo_compute_loss"] +UnslothEfficientGRPO = RL_REPLACEMENTS["UnslothEfficientGRPO"] grpo_accumulated_loss = RL_REPLACEMENTS["grpo_accumulated_loss"] RL_PRE_ITEMS["grpo_trainer"].append(inspect.getsource(grpo_compute_loss)) +RL_PRE_ITEMS["grpo_trainer"].append(inspect.getsource(UnslothEfficientGRPO)) RL_PRE_ITEMS["grpo_trainer"].append(inspect.getsource(grpo_accumulated_loss)) # Edit _get_per_token_logps to handle mixed precision