Guard torch.compile on ROCm when triton_key is missing

[hnxnq7]

Summary

• On ROCm, some Triton builds don’t expose triton_key, which makes the torch.compile/Inductor path crash during LoRA backward.
• This change detects that case and disables torch.compile/Inductor only on ROCm, avoiding the backend crash while leaving CUDA behavior unchanged.

Changes

• unsloth/import_fixes.py: add fix_rocm_triton_key_error() to detect missing triton_key on ROCm and set TORCHINDUCTOR_DISABLE + TORCH_COMPILE_DISABLE.
• unsloth/init.py: invoke the new fix during import.

Behavior

• ROCm only; no change on NVIDIA/CUDA.
• If Triton already has triton_key, nothing is changed.

[gemini-code-assist]

Summary of Changes

Hello @hnxnq7, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request implements a targeted solution to prevent crashes in ROCm environments when using torch.compile or Inductor, which can occur if the installed Triton version lacks the triton_key symbol. By conditionally disabling these features only when the specific problematic condition is met on ROCm, the change ensures stability without affecting CUDA users or ROCm users with compatible Triton builds. This enhances the robustness of the system by gracefully handling a known platform-specific dependency issue.

Highlights

• ROCm Compatibility Fix: Introduces a fix specifically for ROCm environments where torch.compile and Inductor might crash due to a missing triton_key in some Triton builds.
• Conditional Disablement: The system now detects if triton_key is absent on ROCm and, if so, automatically disables torch.compile and Inductor by setting specific environment variables (TORCHINDUCTOR_DISABLE, TORCH_COMPILE_DISABLE).
• No Impact on CUDA: This change is strictly for ROCm and does not alter the behavior or functionality on NVIDIA/CUDA platforms.
• Idempotent Behavior: If Triton already exposes triton_key on ROCm, the fix does nothing, ensuring no unnecessary changes to working configurations.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces a safeguard for ROCm environments where an incomplete Triton installation (missing triton_key) can cause torch.compile to crash. The fix correctly identifies this scenario and disables torch.compile and Inductor by setting environment variables, which is a clean and effective approach. The changes are well-isolated. My review includes a suggestion to improve the robustness of the import checks in the new fix_rocm_triton_key_error function by using more specific exception handling.