Revert "Revert Qwen3 235B GB300 MXFP8 large scale mapping (#2338)"

[ko3n1g]

This reverts commit d3d8030.

What does this PR do ?

Add a one line overview of what this PR aims to accomplish.

Changelog

• Add specific line by line info of high level changes in this PR.

GitHub Actions CI

See the CI sectionin the Contributing doc for how to trigger the CI. A Nvidia developer will need to approve and trigger the CI for external contributors.

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 add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

If you haven't finished some of the above items you can still open "Draft" PR.

Additional Information

• Related to # (issue)

Summary by CodeRabbit

• Chores
  • Streamlined large-scale workload configuration by removing explicit parallelism and CUDA graph scope parameters, simplifying performance tuning settings.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

---

📝 Walkthrough

Walkthrough

A configuration parameter removal from the QWEN3 large-scale pretrain variant, reducing explicit settings for virtual pipeline model parallelism, expert model parallelism, and CUDA graph scoping, while retaining the global batch size override.

Changes

Cohort / File(s)	Summary
QWEN3 Configuration Update scripts/performance/configs/qwen/qwen3_workload_base_configs.py	Removed three parameters (virtual_pipeline_model_parallel_size=12, expert_model_parallel_size=16, cuda_graph_scope=["moe_router", "moe_preprocess"]) from QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_FP8_MX_LARGE_SCALE replace() call, leaving only global_batch_size=512 override.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

Possibly related PRs

• Revert Qwen3 235B GB300 MXFP8 large scale mapping #2338: Directly related—modifies the same config with inverse changes, adding the parallelism and CUDA graph scope settings that this PR removes.
• Ko3n1g/chore/reapply 2152 and 2209 #2273: Related configuration updates to QWEN3 workload configs affecting the same parallelism and CUDA graph parameters across GB300/GB200 variants.

Suggested labels

r0.3.0

Suggested reviewers

• erhoo82
• dingqingy-nv

🚥 Pre-merge checks | ✅ 3 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Test Results For Major Changes	⚠️ Warning	PR removes critical parallelism and CUDA graph configuration parameters from a large-scale training preset without test results, performance metrics, or evidence in the description.	Update PR description with test results, performance comparisons, and explanation of why this revert is necessary and whether it prevents regressions.

✅ Passed checks (3 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the changeset as a revert of a previous revert, specifically relating to Qwen3 235B GB300 MXFP8 large scale mapping configuration.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

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

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch ko3n1g/revert/d3d8030fac4ea69cc228151353c9288acbf2fe6f

---

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.}