Update rl_replacements.py

This commit is contained in:
Daniel Han 2025-02-17 22:30:13 -08:00
commit 6bac0780bc

View file

@ -239,6 +239,7 @@ def grpo_trainer_compute_loss(function_name, function):
ref_per_token_logps, per_token_logps, input_ids, completion_mask, self.beta, advantages, bsz,
)
else:
print(self.args.unsloth_num_chunks, end = ",")
loss, completion_length, mean_kl = grpo_accumulated_loss(
self, _input_ids, logits_to_keep, completion_mask, advantages,
n_chunks = self.args.unsloth_num_chunks,