AWQ CohereForCausalLM & Gemma mappings

[brian-dellabetta]

SUMMARY:
Add AWQ mappings for CohereForCausalLM models, which don't have a post_attention_layernorm and instead run MLP and self_attn computations in parallel

Resolves #1566 (in addition to changes landed since 0.5.1 release)

TODOs:

• Don't land until after 0.6.0 release
• Validate full run on CohereLabs/c4ai-command-r-plus (ran to completion, failed on generate because model couldn't fit on a single H100)

TEST PLAN:
on this branch, AWQ ran on CohereLabs/c4ai-command-r-plus

[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.

[gemini-code-assist]

Summary of Changes

Hello @brian-dellabetta, 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 support for applying Activation-aware Weight Quantization (AWQ) to CohereForCausalLM models. It specifically addresses the unique architectural design of these models, where the Multi-Layer Perceptron (MLP) and self-attention computations run in parallel, by defining appropriate quantization mappings.

Highlights

• AWQ Mappings for Cohere Models: Added new Activation-aware Weight Quantization (AWQ) mappings specifically tailored for CohereForCausalLM models.
• Parallel MLP/Attention Architecture Support: The new mappings are designed to correctly handle Cohere's unique architecture, where the Multi-Layer Perceptron (MLP) and self-attention computations run in parallel, by defining appropriate quantization dependencies.

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

The pull request adds AWQ mappings for CohereForCausalLM models, adapting to their architecture where the input_layernorm feeds both attention and MLP blocks in parallel. The changes are clear and consistent with the codebase. No issues of medium or higher severity were found. Testing on CohereLabs/c4ai-command-r-plus will validate these mappings.

[kylesayrs]

Sweet

[phaelon74]

Thank you @brian-dellabetta for adding these in. You might want to also add:

 "Cohere2ForCausalLM": _cohere_mappings,

As some of the newer Cohere models use this one as well. Your _cohere_mappings work for this as well without issue, but naming wise it wants the 2 in it.

This worked and I was able to quant a 111B model with Four H100s and ~460GB of system RAM.