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 2025-02-19 16:37:12 -08:00
commit 94264f7017
1 changed files with 8 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

9
unsloth/models/rl.py
View file

@ -481,7 +481,7 @@ def _patch_trl_rl_trainers(trainer_file = "grpo_trainer"):
RLTrainer_source,
f"trl.trainer.{trainer_file}",
imports,
overwrite = False,
overwrite = True,
)
# Patch Trainer
@ -547,6 +547,13 @@ def patch_functions(RLTrainer, trainer_file, RLTrainer_name, all_imports, import
)
if len(sampling_params) == 1:
sampling_params = sampling_params[0]
# Fix guided_decoding
sampling_params = sampling_params.replace(
"guided_decoding=guided_decoding,",
'GuidedDecodingParams(backend="outlines", regex=args.vllm_guided_decoding_regex) '\
'if getattr(args, "vllm_guided_decoding_regex", None) is not None else None',
)
# Replace with our vLLM engine
sampling_params = \
" "*12 + "self.llm = model.vllm_engine; self._last_loaded_step = 0; " + \