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 2025-02-14 15:58:13 -08:00
commit b2e003b541
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

@ -248,7 +248,7 @@ RL_FUNCTIONS["grpo_trainer"].append(grpo_trainer_compute_loss)
# https://github.com/huggingface/trl/blob/main/trl/trainer/grpo_trainer.py#L356
# TRL warns if batch size is not a multiple of num_generations -> fix this.
def grpo_trainer_fix_batch_size(RLTrainer_source, RLConfig_source):
if "multiple of num_generations" not in RLTrainer_source: return ""
if "divisible by the number of generations" not in RLTrainer_source: return ""
if "num_generations" not in RLConfig_source: return ""
check_batch_size = \