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 2025-02-17 19:45:07 -08:00
commit fe483b6210
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
unsloth/models/rl_replacements.py
View file

@ -233,7 +233,7 @@ def grpo_trainer_compute_loss(function_name, function):
# loss = ((per_token_loss * completion_mask).sum(dim=1) / completion_mask.sum(dim=1)).mean()
input_ids = input_ids[:, -logits_to_keep:]
loss, completion_length, mean_kl = grpo_compute_loss(
ref_per_token_logps, per_token_logps, input_ids, completion_mask, self.beta, advantages,
ref_per_token_logps, per_token_logps, input_ids, completion_mask, self.beta, advantages, bsz,
)
accumulated_loss, accumulated_completion_length, accumulated_mean_kl = grpo_accumulated_loss(
self, input_ids, logits_to_keep, completion_mask, advantages, n_chunks = 1,