Skip to content

GRPO unwrap fix #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
sidnarayanan wants to merge 6 commits into
base: main
Choose a base branch
from
Open

GRPO unwrap fix #4

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
5e029bf
make sure model is in eval mode before generating
sidnarayanan Jan 29, 2025
7f74ce3
bit more logging
sidnarayanan Jan 29, 2025
61e94cc
switch to transformers logging
sidnarayanan Jan 29, 2025
3bf982c
add more logs
sidnarayanan Feb 2, 2025
edf5b0a
unwrap wrapped model
sidnarayanan Feb 6, 2025
e1063fa
Merge branch 'grpo-unwrap-fix' into grpo-fix-fork
sidnarayanan Feb 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion trl/trainer/grpo_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -500,10 +500,12 @@ def _prepare_inputs(self, inputs: dict[str, Union[torch.Tensor, Any]]) -> dict[s
prompt_completion_ids = torch.cat([prompt_ids, completion_ids], dim=1)
else:
# Regular generation path
with unwrap_model_for_generation(self.model, self.accelerator) as unwrapped_model:
with unwrap_model_for_generation(self.model_wrapped, self.accelerator) as unwrapped_model:
unwrapped_model.eval() # Needed to make sure use_cache works with gradient_checkpointing
prompt_completion_ids = unwrapped_model.generate(
prompt_ids, attention_mask=prompt_mask, generation_config=self.generation_config
)
self.model_wrapped.train()

# Compute prompt length and extract completion ids
prompt_length = prompt_ids.size(1)
Expand Down