Skip to content

Add support of SGD to GRPO trainer #1

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
beryll1um wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add support of SGD to GRPO trainer #1

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
17 changes: 13 additions & 4 deletions src/genrl/trainer/grpo_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,7 @@ class GRPOTrainerConfig:
min_p: float | None = None
repetition_penalty: float = 1.0
num_iterations: int = 1
optimizer: str = "Adam"


class GRPOLanguageTrainerModule(TrainerModule, LoggerMixin):
Expand All @@ -67,9 +68,17 @@ def __init__(self, models: List[Any], config: GRPOTrainerConfig, **kwargs):

self.args = config

self.optimizer = torch.optim.Adam(
self.model.parameters(), lr=self.args.learning_rate
)
match config.optimizer:
case "Adam":
self.optimizer = torch.optim.Adam(
self.model.parameters(), lr=self.args.learning_rate
)
case "SGD":
self.optimizer = torch.optim.SGD(self.model.parameters(),
lr=self.args.learning_rate,
momentum=0.9)
case _:
assert "For GRPO training only SGD or Adam optimizers are supported"

# Tokenizers
self.processing_class = kwargs.get("processing_class", None)
Expand Down Expand Up @@ -446,7 +455,7 @@ def step(

loss.backward()
self.optimizer.step()
self.model.zero_grad()
self.model.zero_grad(set_to_none=True)

metrics = {"train/loss": loss.cpu().mean().item()}
metrics.update({"train/rewards": rewards.cpu().mean().item()})
Expand Down