NVIDIA-NeMo / NVIDIA-NeMo/Gym

Write an RFC for post-training with NOOA agents

Open
#3,459 0 comments 0 reactions 1 assignee View on GitHub

@init-nikhil is already working on this.

Since Sep 17, 2026.

area:training feature needs-design training
Dominant language
Python
Stars
1.2k
Forks
349
Avg merge
1d 23h
Merged PRs (30d)
324

Description

Parent

Sub-issue of #3458.

Goal

Write an RFC defining how NOOA agents participate in Gym-based post-training on top of #2096 and the CLU evaluation work in #3453.

Coordinate with Cynthia and whoever Cynthia points to for testing out the post-training readiness at scale.

Required decisions

  • Define the training-ready rollout and trajectory contract.
  • Define token ID, log probability, model-call lineage, reward, and failure requirements.
  • Define how NOOA policy calls are routed through Gym's model server during collection.
  • Define state isolation, retries, cancellation, and reproducibility requirements.
  • Identify required changes in NOOA, Gym, CLU, and the selected training framework.
  • Define staged implementation milestones and end-to-end readiness tests.

Acceptance criteria

  • The RFC documents the current state and remaining gaps.
  • Data ownership and integration boundaries are explicit.
  • Evaluation-only and post-training requirements are distinguished.
  • The proposal includes alternatives, risks, and unresolved decisions.
  • The proposal includes measurable milestones for implementation and validation.
  • Relevant NOOA, Gym, CLU, and training stakeholders review the RFC.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.