ci(fix): PerfPlugin for llama

[ko3n1g]

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

• Refactor
  • Improved model identification and environment variable configuration logic during pretraining phase, enhancing consistency and handling of Llama model variants across different sizes and configurations.
  • Standardized hardware-specific performance optimizations across multiple systems (h100, gb200, gb300), ensuring more predictable and consistent training behavior with enhanced model classification logic and optimized configuration management for better results.

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

[coderabbitai]

📝 Walkthrough

Walkthrough

The change refines model-specific environment variable gating conditions in the performance plugins module. It updates checks for Llama-based models by changing model_family_name comparisons from specific subversions (llama31, llama3) to a unified "llama" check, affecting when hardware-specific NCCL and cuDNN optimizations are applied during pretraining.

Changes

Cohort / File(s)	Change Summary
Environment variable gating refinement scripts/performance/perf_plugins.py	Updated model-specific condition checks to use unified model_family_name == "llama" instead of version-specific checks (llama31, llama3\*), affecting applicability of gb200, h100, and gb300 hardware optimizations (NCCL_CTA_POLICY, del_cudnn_ln) during pretraining

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~5 minutes

🚥 Pre-merge checks | ✅ 3 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Test Results For Major Changes	⚠️ Warning	PR alters performance-related gating logic for llama-based models without before/after benchmarks or testing details to demonstrate no regression.	Add detailed performance results comparing previous and updated behavior with hardware configuration, model, dataset, and measurement methodology.

✅ 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 'ci(fix): PerfPlugin for llama' is directly related to the changeset, which refines llama model-specific environment variable gating in the PerfPlugin by updating model checks from 'llama31' to 'llama' family patterns.
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

---

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.}