ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update rl_replacements.py

Browse source
This commit is contained in:
Daniel Han 2025-02-16 16:14:21 -08:00
commit 63ed4d8d74
1 changed files with 1 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

1
unsloth/models/rl_replacements.py
View file

@ -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,
)