NVIDIA-NeMo / NVIDIA-NeMo/Gym

CLU evaluations for NOOA agents

Open
#3,453 0 comments 0 reactions 1 assignee View on GitHub

@init-nikhil is already working on this.

Since Sep 17, 2026.

area:agent area:evaluation dependency-CLU tracking
Dominant language
Python
Stars
1.2k
Forks
349
Avg merge
1d 23h
Merged PRs (30d)
324

Description

Goal

Run NOOA agents through the CLU evaluation workflow and produce reproducible Nemotron 3.5 results on DeepSWE and SWE Bench Pro.

Scope

  • Implement a sandboxed NOOA runner suitable for CLU execution.
  • Integrate NOOA with DeepSWE.
  • Integrate NOOA with SWE Bench Pro.
  • Produce evaluation results for NOOA agents with Nemotron 3.5.
  • Analyze benchmark traces and implement concrete NOOA harness optimizations.

Features

  • #3456
  • #3454
  • #3455
  • #3457
  • #3506

Success criteria

  • NOOA evaluations can be launched through the expected CLU workflow.
  • Sandboxed runs preserve rollout isolation and produce usable Gym artifacts.
  • DeepSWE and SWE Bench Pro complete through their canonical verification paths.
  • Nemotron 3.5 results are recorded with reproducible configuration, rollout artifacts, rewards, completion rates, and categorized failures.
  • Trace analysis produces validated NOOA harness optimizations for DeepSWE and SWE Bench Pro.

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.