docs: V0.5 perf results

[guyueh1]

What does this PR do ?

Update performance results for v0.5

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

  • Expanded performance summary with new benchmark sections covering H100 BF16, H100 FP8, and GB200 BF16 configurations with detailed results tables.

• New Features

  • Added GRPO performance testing configuration for Qwen3-30B with 128K context length support.
  • Added automated performance test suite for GRPO experiments including metrics validation.

• Chores

  • Updated submodule dependency references.

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

[github-actions]

✅ Submodule Fast-Forward Check Results

Check based on commit: dd8189a (PR #1771 from v0.5_perf_results)

✅ Submodules that are properly updated:

Megatron-Bridge: ✅ PR branch is ahead of main branch (fast-forward)

All submodule changes look good! ✨

[coderabbitai]

Caution

Review failed

The pull request is closed.

📝 Walkthrough

Walkthrough

This PR updates the Megatron-Bridge submodule, expands the performance benchmark documentation with H100/GB200 results, introduces a new GRPO configuration for Qwen3-30B with 128K sequence length, adds rope-scaling (yarn) embedding support in model imports, and registers a new performance test.

Changes

Cohort / File(s)	Summary
Submodule Update 3rdparty/Megatron-Bridge-workspace/Megatron-Bridge	Pointer advanced from 1e9a459 to 812dcf75; no functional changes.
Documentation docs/about/performance-summary.md	Restructured and expanded with three new benchmark sections (H100 BF16, H100 FP8, GB200 BF16), introducing explicit Algorithm and Model columns, updated metrics layout, and dataset references.
GRPO Qwen3-30B 128K Support examples/configs/recipes/llm/performance/grpo-qwen3-30ba3b-4n8g-128K.yaml	New configuration enabling 128K sequence length with tensor/pipeline/expert/context parallelism, yarn rope scaling, and logging setup.
Model Import Enhancement nemo_rl/models/megatron/community_import.py	Added conditional logic to apply yarn-based rope scaling embeddings when NRL_MCORE_OVERRIDE_EMBEDDING_TYPE=1 and rope_type="yarn" in config overrides.
Performance Test Suite tests/test_suites/llm/performance/grpo-qwen3-30ba3b-4n8g-128K.sh	New shell script executing GRPO performance experiment with metric validation for training loss and token probability error thresholds.
Test Registry tests/test_suites/performance_h100.txt	Added new test script path to SYNC suite.

Estimated Code Review Effort

🎯 3 (Moderate) | ⏱️ ~22 minutes

Possibly Related PRs

• cp: feat: Onboard perf recipes in tests (1322) into r0.4.0 #1497: Adds LLM performance recipes and test scripts for GRPO experiments with matching structure and test suite registrations.
• feat: Support for nano-v2 #1514: Modifies nemo_rl/models/megatron/community_import.py for Megatron import path/config handling.
• chore: Update RL to use megatron-bridge tot #1358: Updates import_model_from_hf_name function in the same module for model provider initialization.

Suggested Labels

documentation, CI:docs, r0.5.0

Suggested Reviewers

• terrykong

✨ Finishing touches

• 📝 Generate docstrings

---

📜 Recent 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 dd8189a.

📒 Files selected for processing (6)

• 3rdparty/Megatron-Bridge-workspace/Megatron-Bridge
• docs/about/performance-summary.md
• examples/configs/recipes/llm/performance/grpo-qwen3-30ba3b-4n8g-128K.yaml
• nemo_rl/models/megatron/community_import.py
• tests/test_suites/llm/performance/grpo-qwen3-30ba3b-4n8g-128K.sh
• tests/test_suites/performance_h100.txt

---

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.}