Fix tinker loss device mismatch#115
Conversation
Summary of ChangesHello, 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 addresses a critical bug within the Tinker compatibility layer that caused device mismatch errors during loss computation. By explicitly moving Highlights
🧠 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. Changelog
Using Gemini Code AssistThe 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
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 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
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request addresses a device mismatch bug in the Tinker compatibility path when computing elementwise_loss. The fix involves moving the token_log_probs tensor to the same device as weights before multiplication, which correctly resolves the issue. I have added a suggestion to further improve performance by moving the device transfer operation out of the loop, which would make the change in this pull request unnecessary.
There was a problem hiding this comment.
Pull request overview
Fixes a device mismatch in the Tinker compatibility layer during forward_backward by ensuring per-token log probabilities and loss weights live on the same device before computing elementwise loss.
Changes:
- Move
token_log_probsontoweights.deviceprior toelementwise_loss = -token_log_probs * weights.
You can also share your feedback on Copilot code review. Take the survey.
PR type
PR information
This PR fixes a device mismatch issue in the Tinker compatibility path during
forward_backward.Previously, in
src/twinkle/server/tinker/common/compat_base.py,_get_forward_output()could mix tensors from different devices when computing:In the Tinker path, logps may be materialized on CPU for transport, while weights and related tensors are placed on the accelerator device. This could trigger errors like:
Fix: