verl-project/verl

Support Generative Reward Model (GenRM)

Open

#229 opened on Feb 9, 2025

View on GitHub
 (17 comments) (0 reactions) (1 assignee)Python (21,533 stars) (3,940 forks)auto 404
enhancementgood first issue

Description

According to the documentation, veRL only supports AutoModelForSequenceClassification. What would be the best way to implement generative reward model (GenRM) for veRL? I tried looking at FSDP Workers and Megatron-LM Workers but they no longer exist.

Contributor guide

Tech stack
pythonpytorch
Domain
machine learningai
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
5
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
over 1 week
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
needs maintainer response
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
understanding of RLHFfamiliarity with verl architectureknowledge of generative reward modelsPython and PyTorch
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
15
Research direction
Investigate the current reward model implementation in verl, specifically the AutoModelForSequenceClassification support. Review the comments and any linked PRs for proposed designs. Since the referenced workers no longer exist, explore the current actor/critic architecture to identify where GenRM integration would fit. Propose a design that extends the reward model to support generation based scoring.