Skip to content

Add grpo difficulty curriculum#1694

Open
undfined wants to merge 21 commits into
split-difficulty-curriculum-corefrom
split-grpo-difficulty-curriculum
Open

Add grpo difficulty curriculum#1694
undfined wants to merge 21 commits into
split-difficulty-curriculum-corefrom
split-grpo-difficulty-curriculum

Conversation

@undfined
Copy link
Copy Markdown

No description provided.

Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request implements a difficulty-aware curriculum for GRPO training. It introduces a new DifficultyCurriculumHFDataLoader that samples prompts based on difficulty metadata, a utility script create_difficulty_map.py to generate this metadata from pass-rate datasets, and several training configuration scripts. The reviewer provided actionable feedback, including a fix for epoch-based index repetition in the dataloader, a correction for NumPy type serialization in the difficulty map script, and a fix for a typo in a dataset repository path.

Comment thread open_instruct/data_loader.py Outdated
Comment thread scripts/data/difficulty_sampling/create_difficulty_map.py Outdated
@allenai allenai deleted a comment from gemini-code-assist Bot May 13, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant