Nice Work on CodeRM but Many Missing Citations

[ashwin296]

This is a really nice work that contributes to new reward models in the coding domain.

However, many relevant prior works are not cited. The claim that "very few models have explored the potential of reinforcement learning" in the introduction is misleading (currently, only CodeRL is cited in Section 5.4).

To name a few:

• PPOCoder
• RLEF
• RLTF
• StepCoder
• DSTC
• Bellman Coder
• CodeGemma
• ...

It would be great to clarify the contributions to reward model training (which are quite valuable) and adjust the phrasing to better reflect prior works.

Thanks again for open-sourcing!

-- a Reinforcement Learner : )

[wenhuchen]

Thanks for the reminder. We have read these insightful papers. We will compare with them in the paper.

[ashwin296]

And PG-TD, where there are value estimates.