[WIP] [Research] Attention quantization and transformation

[kylesayrs]

neuralmagic/compressed-tensors#374

[gemini-code-assist]

Summary of Changes

Hello @kylesayrs, 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 focuses on implementing attention quantization and transformation. It introduces a new calibrated_attention function that integrates quantization and transformation steps into the attention mechanism. The changes also include refactoring the calibration process by removing KV cache related hooks and modifying observer initialization to handle different module types correctly. Additionally, the attention interface is wrapped to dynamically switch the attention implementation during calibration.

Highlights

• Attention Quantization: Introduces a calibrated_attention function that integrates quantization and transformation steps into the attention mechanism. This includes applying transforms to query and key tensors, calibrating quantization, and applying forward quantization.
• Calibration Hooks: Removes KV cache related calibration hooks and initialization functions, streamlining the calibration process. Also, the calibration hooks are now initialized per module instead of the entire model.
• Observer Initialization: Modifies the observer initialization process to handle different module types (Linear, Embedding, and Attention) separately, ensuring correct observer setup for each.
• Attention Implementation Wrapping: Wraps the attention interface to dynamically switch the attention implementation to calibrated_attention during calibration, ensuring that the custom attention function is used.

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 in your pull request via creating an issue comment (i.e. comment on the pull request page) using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands.

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 issue 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 is currently in preview and 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 to provide feedback.

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 new approach for attention quantization by refactoring the existing logic. It adds a calibrated_attention function to handle quantization of query, key, and value tensors, and removes the previous KV cache-specific implementation. The changes are a good step towards a more flexible quantization framework.

However, my review has identified several critical and high-severity issues that must be addressed. These include:

• A resource leak where calibration hooks are not being properly removed.
• A copy-paste error in the new attention calibration logic, which would lead to incorrect model calibration.
• A logic bug that would prevent output activation quantization from being applied in certain scenarios.
• A commented-out section in the new attention function that contains a bug and represents a missing feature.

I've provided detailed comments and suggestions for each of these issues. Addressing them will be crucial for the stability and correctness of the new quantization mechanism.

[github-actions]

👋 Hi! Thank you for contributing to llm-compressor. Please add the ready label when the PR is ready for review.

Note: This is required to complete the testing suite, please only add the label once the PR is code complete and local testing has been performed.