docs: v0.5 performance results update

[guyueh1]

What does this PR do ?

As title

Issues

List issues that this PR closes (syntax):

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

• ...

Summary by CodeRabbit

• Documentation
  • Added new performance benchmark sections for H100 BF16, H100 FP8, and GB200 BF16 configurations with system details and dataset references
  • Restructured benchmark tables for improved clarity with new Algorithm column
  • Updated benchmark metrics and performance data across all sections

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[coderabbitai]

📝 Walkthrough

Walkthrough

Documentation update to the performance summary benchmark file. Adds three new benchmark sections for H100 BF16/FP8 and GB200 BF16 configurations with restructured table format including an Algorithm column and updated metrics for training and generation parameters.

Changes

Cohort / File(s)	Summary
Performance Summary Documentation docs/about/performance-summary.md	Adds three new benchmark sections (H100 BF16, H100 FP8, GB200 BF16) with expanded table structure; replaces compact header with Algorithm-inclusive format; updates metric rows (T-Max Length, G-Average Seq len, #GPUs, G-GBS, T-GBS, Tokens/sec per GPU, Total Step time) across GRPO and DAPO algorithm entries.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Possibly related PRs

• docs: Create performance-summary.md for NeMo RL #1560: Directly expands and restructures the same performance-summary.md file by adding H100/GB200 benchmark sections and implementing the new table format with algorithm column introduced in this PR.

Suggested reviewers

• terrykong

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'docs: v0.5 performance results update' directly and clearly reflects the main change: updating documentation with v0.5 performance results as shown in the raw summary.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Test Results For Major Changes	✅ Passed	This pull request contains only minor changes—it updates performance benchmark documentation without introducing any code modifications, new features, breaking changes, or significant refactoring.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Fix all issues with AI agents

In @docs/about/performance-summary.md:
- Line 46: The section header "## Nemo RL v0.4" is inconsistent with the PR
title referencing v0.5; update the header text in the markdown (replace "Nemo RL
v0.4" with "Nemo RL v0.5") so the documentation version matches the PR and other
references to the release.

📜 Review details

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between e95efb9 and 6babe73.

📒 Files selected for processing (1)

• docs/about/performance-summary.md

🧰 Additional context used

📓 Path-based instructions (2)

docs/**/*.md

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

Update docs/index.md when a new markdown doc is added under docs/**/*.md or a markdown file is renamed, ensuring the document appears in the most appropriate section

Files:

• docs/about/performance-summary.md

!(**/tests/**|**/test_*.py|**/test_*.sh)

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

Add the NVIDIA copyright header to all Python files and shell scripts (excluding tests). The header should include the current year

Files:

• docs/about/performance-summary.md

🧠 Learnings (1)

📚 Learning: 2025-11-24T17:24:47.707Z

Learnt from: CR
Repo: NVIDIA-NeMo/RL PR: 0
File: coderabbit-custom-pre-merge-checks-unique-id-file-non-traceable-F7F2B60C-1728-4C9A-8889-4F2235E186CA.txt:0-0
Timestamp: 2025-11-24T17:24:47.707Z
Learning: If a change could affect performance, the PR description should include before-and-after performance numbers, as well as the configuration and context in which they apply

Applied to files:

• docs/about/performance-summary.md

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (6)

• GitHub Check: build-container / main
• GitHub Check: Lint check
• GitHub Check: build-container / main
• GitHub Check: Lint check
• GitHub Check: Post submodule check comment / Comment on PR
• GitHub Check: Post automodel integration comment / Comment on PR

🔇 Additional comments (2)

docs/about/performance-summary.md (2)

48-78: LGTM! Well-structured benchmark sections.

The H100 BF16 and FP8 benchmark sections are well-organized with clear metadata and properly formatted tables. The addition of the "Algorithm" column effectively distinguishes between GRPO and DAPO results, and the dataset references on line 49 appropriately mention both algorithms used.

---

79-96: LGTM! GB200 benchmark section is well-structured.

The GB200 BF16 benchmark section follows the same clear structure as the H100 sections with appropriate system metadata and properly formatted tables. The results showcase performance on the new GB200-NVL72 system.

[guyueh1]

@snowmanwwg please take a look at results
https://github.com/guyueh1/NeMo-RL/blob/6babe73584df94bd507b39f69d2ec73f86311a86/docs/about/performance-summary.md

[terrykong]

@guyueh1 could you link the performance yaml recipes dir as part of the docs? would help folks find out how to reproduce this on their infra

[guyueh1]

@guyueh1 could you link the performance yaml recipes dir as part of the docs? would help folks find out how to reproduce this on their infra

Added a link to the recipes in r0.5.0 branch