Patch in tiled mlp

[mmathew23]

Depends on unslothai/unsloth-zoo#350

This adds the check for environment variable UNSLOTH_TILED_MLP to enable tiled_mlp via patch.

It also adjusts the patching logic for the llama style models to account for the lora mlp kernels with tiled_mlp.

[gemini-code-assist]

Summary of Changes

Hello @mmathew23, 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 integrates a 'tiled MLP' optimization into the Unsloth framework, designed to enhance performance for certain model architectures. It provides a mechanism for users to activate this feature through an environment variable, and crucially, it refines the interaction between LoRA MLP kernels and the new tiled MLP implementation in Llama-style models to maintain correct behavior and efficiency.

Highlights

• Tiled MLP Integration: Introduces the ability to enable 'tiled MLP' functionality within the framework, which is a performance optimization for neural networks.
• Environment Variable Control: The tiled MLP feature is controlled via the UNSLOTH_TILED_MLP environment variable, allowing users to activate it with specific options like 'arctic' or 'target'.
• LoRA MLP Kernel Adjustment: Adjusts the patching logic for Llama-style models to correctly account for LoRA MLP kernels when the tiled MLP optimization is active, ensuring compatibility and proper function.

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 adds support for Tiled MLP patching, controlled by the UNSLOTH_TILED_MLP environment variable. The changes in unsloth/models/llama.py correctly adjust the LoRA patching logic for this. However, I've identified a critical issue in unsloth/models/llama.py where a method is being patched incorrectly, which will likely lead to a TypeError. Additionally, the code to apply the Tiled MLP patch in unsloth/models/loader.py is duplicated, and I've recommended refactoring this to improve maintainability.

[gemini-code-assist]

This block of code for patching Tiled MLP is duplicated in FastModel.from_pretrained (lines 1149-1153). To improve maintainability and avoid code duplication, consider extracting this logic into a private helper function within this module and calling it from both from_pretrained methods.

[gemini-code-assist]

This is a duplicate of the Tiled MLP patching logic found in FastLanguageModel.from_pretrained (lines 571-575). This should be refactored into a shared helper function to avoid code duplication.