Skip to content

Add GRPO full finetune single device #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
yxjiang wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add GRPO full finetune single device #1

Show file tree
Hide file tree
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
4 changes: 2 additions & 2 deletions recipes/dev/grpo_full_finetune_distributed.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,7 @@
from torchtune.config._utils import _get_component_from_path
from torchtune.datasets import ConcatDataset
from torchtune.dev.rl.generation import generate
from torchtune.dev.rl.rewards import batched_rewards
from torchtune.dev.grpo.rewards import batch_shaped_correctness_reward
from torchtune.dev.rl.types import GRPOStats, GRPOTrajectory
from torchtune.modules import local_kv_cache
from torchtune.recipe_interfaces import FTRecipeInterface
Expand Down Expand Up @@ -646,7 +646,7 @@ def generate_trajectory(
# Do some reward modelingggggggg
# responses :: [B x G, L]
responses = responses.reshape(batch_size, grpo_size, -1) # [B, G, L]
rewards, successes = batched_rewards(self._tokenizer, responses, answers)
rewards, successes = batch_shaped_correctness_reward(self._tokenizer, responses, answers)
rewards = rewards.to(self._device) # [B, G]
successes = successes.to(self._device) # [B, G]

Expand Down
Loading