From 63ed4d8d747991ccc9079accf0bafdc51efb00e2 Mon Sep 17 00:00:00 2001 From: Daniel Han Date: Sun, 16 Feb 2025 16:14:21 -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 ad6d0f2bbc..ad6d7822ac 100644 --- a/unsloth/models/rl_replacements.py +++ b/unsloth/models/rl_replacements.py @@ -229,6 +229,7 @@ def grpo_trainer_compute_loss(function_name, function): # per_token_loss = -(per_token_loss - self.beta * per_token_kl) # loss = ((per_token_loss * completion_mask).sum(dim=1) / completion_mask.sum(dim=1)).mean() input_ids = input_ids[:, -logits_to_keep:] + print(input_ids.shape, ref_per_token_logps.shape, per_token_logps.shape, completion_mask.shape, advantages.shape) loss, completion_length, mean_kl = grpo_compute_loss( ref_per_token_logps, per_token_logps, input_ids, completion_mask, self.beta, advantages, )