Update rl_replacements.py

This commit is contained in:
Daniel Han 2025-02-17 20:21:20 -08:00
commit 673325dea7

View file

@ -237,7 +237,7 @@ def grpo_trainer_compute_loss(function_name, function):
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,
self, _input_ids, logits_to_keep, completion_mask, advantages, n_chunks = 1,
)
print("loss", loss, accumulated_loss)
print("completion_length", completion_length, accumulated_completion_length)