Open Issues Need Help
View All on GitHub Add an option to save checkpoints in HF format about 1 month ago
enhancement good first issue
Remove unused converter_type 2 months ago
good first issue t-mcore
Importing dataset classes defined outside of NeMo-RL 2 months ago
good first issue data module
Make `only_unmask_final` flag configurable for SFT. 3 months ago
enhancement good first issue community-request waiting-on-maintainers
Create RL playbook as a follow up to DAPT 3 months ago
enhancement good first issue
More helpful error message if prepare_for_*() not called 11 months ago
good first issue UX
[Chore] Clean up config defaults throughout codebase 12 months ago
AI Summary: The task involves cleaning up default values in the configuration files of a reinforcement learning library (NeMo RL) to adhere to a specific design philosophy. This requires reviewing several Python files, identifying instances where default values are hardcoded, and refactoring the code to use configuration files instead. The goal is to improve maintainability and consistency.
Complexity:
4/5
good first issue
Support cons@k in eval about 1 year ago
good first issue
Support pass@k in eval about 1 year ago
good first issue