feat(attention): implement lightweight MultiheadAttention using PyTorch SDPA

[StatMixedML]

• Uses PyTorch scaled_dot_product_attention with Flash Attention dispatch
• More efficient than naive attention; allows for longer sequences (larger T) or bigger batches before OOM

[gemini-code-assist]

Summary of Changes

Hello @StatMixedML, 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 introduces a new, highly optimized MultiheadAttention module that leverages PyTorch's scaled_dot_product_attention (SDPA) functionality. This change significantly enhances the efficiency of attention computations, particularly by enabling automatic dispatch to Flash Attention kernels on supported hardware. The integration of this new module into the NanoTabPFNModel is expected to reduce memory consumption and allow for processing longer sequences or larger batches, thereby improving the overall scalability and performance of the model.

Highlights

• New MultiheadAttention Implementation: A new MultiheadAttention module has been introduced, designed to be lightweight and efficient, residing in its own dedicated file nanotabpfn/attention.py.
• PyTorch SDPA Integration: The core of the new attention module leverages PyTorch's scaled_dot_product_attention (SDPA) for its computations, simplifying the implementation and benefiting from PyTorch's optimizations.
• Flash Attention Dispatch: The use of SDPA enables automatic dispatch to highly optimized Flash Attention kernels on compatible CUDA hardware, leading to significant performance improvements and a reduced memory footprint, especially for longer sequences.
• Model Integration and Compatibility: The existing NanoTabPFNModel has been updated to utilize this new, more efficient MultiheadAttention implementation, and the logic for loading model state dictionaries has been adjusted to correctly identify the parameters of the new attention module.

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 lightweight MultiheadAttention module using PyTorch's efficient scaled_dot_product_attention, which is a fantastic performance enhancement. The related changes in interface.py and model.py are correct and necessary adaptations. My feedback is focused on the new nanotabpfn/attention.py file, with suggestions to enhance its flexibility and reusability by exposing parameters for dropout, attention masking, and causality, which would align it more closely with the standard torch.nn.MultiheadAttention module. I've also included a minor suggestion to improve type hint conventions.