NVIDIA-NeMo / NVIDIA-NeMo/Megatron-Bridge

[bug] [Determinism] determinism break of ultra v3 on large scale

Open
#4,906 0 comments 0 reactions 2 assignees View on GitHub

@Connor-XY is already working on this.

Since Jul 16, 2026.

area:model bug Determinism tracking
Dominant language
Python
Stars
920
Forks
505
Avg merge
1d 16h
Merged PRs (30d)
253

Description

Problem

24 nodes (96 GPUs) -> 768 nodes (3072 gpus)
Non-deterministic scaling: bit-exact ✓ -> diverges from iteration 1 ✗

Need to narrow down the root cause of determinism failure.

Minimal repro
follow: https://github.com/NVIDIA-NeMo/Megatron-Bridge/tree/zhiyul/nemotron-3-ultra-perf-recipe
Expected behavior

Determinism stands in large scale like 3k gpu run.

Affected area

area:model

Regression?

Yes

Environment

No response

Logs

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.