disable CG for 8B SFT

[malay-nagda]

What does this PR do ?

disable CG for 8B SFT as it reduces performance.

Changelog

cuda_graph_impl="none",  # NOTE: CUDA Graphs reduces performance here

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

Release Notes

• Chores
  • Updated performance configuration settings for Llama3 8B supervised finetuning workflows to enhance system stability and compatibility across different GPU configurations.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between e3365fe and d25de21.

📒 Files selected for processing (1)

• scripts/performance/configs/llama/llama3_workload_base_configs.py

---

📝 Walkthrough

Walkthrough

Two Llama3 8B SFT configurations in the performance workload configs have CUDA Graphs disabled by changing cuda_graph_impl from "transformer_engine" to "none" in both _LLAMA3_8B_SFT_CONFIG_GB200 and _LLAMA3_8B_SFT_CONFIG_H100_FP8_CS_V1 configurations.

Changes

Cohort / File(s)	Summary
Llama3 CUDA Graphs Configuration scripts/performance/configs/llama/llama3_workload_base_configs.py	Disabled CUDA Graphs in two Llama3 8B SFT configurations by setting cuda_graph_impl from "transformer_engine" to "none" with clarifying comments.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Suggested reviewers

• erhoo82
• ko3n1g

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'disable CG for 8B SFT' directly and specifically describes the main change: disabling CUDA Graphs for 8B SFT configurations.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Test Results For Major Changes	✅ Passed	PR makes minor configuration changes (2 lines) disabling CUDA Graphs in SFT configs, meeting the custom check pass condition for minor changes.

_{✏️ 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 malay/no_CG_8B_SFT_260201

---

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.}