Update Deepseek V3 MXFP8 GB200 mapping

[dingqingy-nv]

What does this PR do ?

Small CG change to achieve 1040 TFLOPs on DSV3 MXFP8 GB200 recipe.

Summary by CodeRabbit

• Chores
  • Updated GPU performance configuration to optimize computation graph scope with additional execution components.

[coderabbitai]

📝 Walkthrough

Walkthrough

Updated the GB200 V1 pretrain configuration to expand cuda_graph_scope from an empty list to include "attn", "moe_router", and "moe_preprocess". This is a single-line configuration modification with no changes to control flow or error handling.

Changes

Cohort / File(s)	Summary
CUDA Graph Scope Configuration scripts/performance/configs/deepseek/deepseek_workload_base_configs.py	Updated GB200 V1 pretrain config to add "attn" to cuda_graph_scope alongside existing "moe_router" and "moe_preprocess" entries.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

Possibly related PRs

• Update Qwen3 235B A22B MXFP8 GB200/300 recipe and resolve NaN grad norm #2209: Adds identical cuda_graph_scope entries to multiple model pretrain configurations.
• Dsv3 Recipe Update #2152: Modifies the same deepseek_workload_base_configs.py file for GB300 V1 variant CUDA graph scope configuration.
• cp: Dsv3 Recipe Update (2152) into r0.3.0 #2186: Updates cuda_graph_scope entries in the same DeepSeek configuration file for a different pretrain variant.

Suggested reviewers

• erhoo82
• ko3n1g

🚥 Pre-merge checks | ✅ 3 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Test Results For Major Changes	⚠️ Warning	PR description mentions intended performance goal of 1040 TFLOPs but lacks actual before-and-after performance metrics validating the CUDA graph scope configuration change.	Add performance test results showing baseline metrics, post-change metrics confirming 1040 TFLOPs achievement, testing environment details, and regression testing results.

✅ 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 mentions 'Deepseek V3 MXFP8 GB200 mapping' which aligns with the configuration change to GB200 V1 pretrain config, but is less specific than the actual change (adding 'attn' to cuda_graph_scope for performance optimization).
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

---

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.}