Revert Qwen3 235B GB300 MXFP8 large scale mapping

[dingqingy-nv]

What does this PR do ?

Revert mapping for perf regression.

Summary by CodeRabbit

• Chores
  • Added a new large-scale configuration variant with optimized parallelism layout and enhanced CUDA graph scoping for performance workloads.

[coderabbitai]

📝 Walkthrough

Walkthrough

Added a new large-scale Qwen3 235B A22B FP8 MX configuration variant (QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_FP8_MX_LARGE_SCALE) with custom settings for virtual pipeline parallelism, expert model parallelism, CUDA graph scoping, and batch size.

Changes

Cohort / File(s)	Summary
Qwen3 Large-Scale Config scripts/performance/configs/qwen/qwen3_workload_base_configs.py	Added new exported configuration constant extending QWEN3_235B_A22B_PRETRAIN_CONFIG_GB300_FP8_MX_V2 with virtual_pipeline_model_parallel_size=12, expert_model_parallel_size=16, cuda_graph_scope=["moe_router", "moe_preprocess"], and global_batch_size=512 for large-scale variant.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~5 minutes

Possibly related PRs

• Update Qwen3 235B A22B MXFP8 GB200/300 recipe and resolve NaN grad norm #2209: Modifies the same Qwen3 GB300 FP8 configuration with potentially conflicting changes to parallelism and cuda_graph_scope parameters.

Suggested labels

performance

Suggested reviewers

• ko3n1g
• erhoo82

🚥 Pre-merge checks | ✅ 3 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Test Results For Major Changes	⚠️ Warning	PR addresses performance regression but lacks before-and-after metrics, configuration context, and validation data required by custom check guidelines.	Add before-and-after performance numbers, hardware configuration details, and benchmarking data validating that the revert resolves the performance regression.

✅ 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 'Revert Qwen3 235B GB300 MXFP8 large scale mapping' accurately describes the main change: adding a reverted configuration for large-scale Qwen3 235B FP8 MX variant due to performance regression.
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

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

No actionable comments were generated in the recent review. 🎉

Tip

Issue Planner is now in beta. Read the docs and try it out! Share your feedback on Discord.

---

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.}