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

Update rl.py

Browse source
This commit is contained in:
Daniel Han 2025-02-17 23:57:57 -08:00
commit 3b5c0d3c74
1 changed files with 7 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

8
unsloth/models/rl.py
View file

@ -442,6 +442,12 @@ def _patch_trl_rl_trainers(trainer_file = "grpo_trainer"):
# Selective log softmax
selective_log_softmax_code = inspect.getsource(selective_log_softmax)
# Trainer kwargs
comma = "" if RLTrainer_call_args.endswith(",") else ","
unsloth_extra_args = comma + \
"vllm_sampling_params = vllm_sampling_params,\n"\
"unsloth_num_chunks = unsloth_num_chunks, **kwargs"
# Get final source code
RLTrainer_source = RLTrainer_replacement.format(
RLTrainer_name = RLTrainer_name,
@ -449,7 +455,7 @@ def _patch_trl_rl_trainers(trainer_file = "grpo_trainer"):
RLTrainer_arguments = RLTrainer_arguments,
RLTrainer_extra_args = RLTrainer_extra_args,
RLTrainer_call_args = RLTrainer_call_args,
RLTrainer_kwargs = ",**kwargs"[1 if RLTrainer_call_args.endswith(",") else 0:],
RLTrainer_kwargs = unsloth_extra_args,
RLConfig_name = RLConfig_name,
__RLConfig_doc__ = __RLConfig_doc__,