Update Qwen3 30B H100 Base Configs with HybridEP

[rhmukundan]

Summary by CodeRabbit

• Chores
  • Optimized Qwen3 model training configurations to enhance performance efficiency across different parallelism strategies.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

---

📝 Walkthrough

Walkthrough

Configuration updates for Qwen3 pretraining on H100/GB300: switching MoE token dispatcher type to "flex" in qwen3_llm_pretrain.py, and adjusting parallelism strategy from pipeline-virtual to expert-model-parallel in qwen3_workload_base_configs.py, with corresponding dispatcher backend and overlap settings changes.

Changes

Cohort / File(s)	Summary
Qwen3 MoE Token Dispatcher Type scripts/performance/configs/qwen/qwen3_llm_pretrain.py	Changed moe_token_dispatcher_type from "alltoall" to "flex" for qwen3_30b_a3b_pretrain_config_h100.
Qwen3 Workload Parallelism and MoE Backend scripts/performance/configs/qwen/qwen3_workload_base_configs.py	Replaced pipeline/virtual parallelism settings with expert_model_parallel_size=16, disabled moe_a2a_overlap, and switched moe_flex_dispatcher_backend from "deepep" to "hybridep" across Qwen3 30B A3B config blocks.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

Possibly related PRs

• fix: perf configs 2 #2393: Sets moe_token_dispatcher_type to "flex" for Qwen3 30B pretraining configs, directly overlapping with qwen3_llm_pretrain.py changes.
• cp: Update Qwen3 235B A22B MXFP8 GB200/300 recipe and resolve NaN grad norm (2209) into r0.3.0 #2210: Modifies qwen3_workload_base_configs.py with identical model-parallel settings changes (removing virtual pipeline sizing and adjusting expert_model_parallel_size).
• qwen gbs 2x #2280: Updates the same QWEN3_30B_A3B pretrain config blocks in qwen3_workload_base_configs.py, causing potential interaction with this PR's changes.

Suggested reviewers

• ko3n1g
• yaoyu-33

🚥 Pre-merge checks | ✅ 3 | ❌ 1

❌ Failed checks (1 inconclusive)

Check name	Status	Explanation	Resolution
Test Results For Major Changes	❓ Inconclusive	PR makes MoE dispatcher configuration changes but test results/performance validation documentation is not accessible in repository context.	Provide PR description from #2477 including any test results, performance benchmarks, or convergence validation documentation for dispatcher backend and parallelism strategy changes.

✅ 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 clearly and specifically summarizes the main change: updating Qwen3 30B H100 configuration with HybridEP dispatcher backend, which is the primary modification across both configuration files.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ 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 rmukundan/qwen3_30b_h100_config_update

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.}

[ko3n1g]

Can you update golden values of the internal CI?