Skip to content

Feasibility of Replacing Transformer with Mamba #57

Description

@linjing-lab

Hi, I have recently been investigating methods used in protein language models and have identified Flow Matching and Mamba as two innovative approaches that could benefit this field. I am interested in understanding whether SimpleFold, described as the first flow-matching-based protein folding model built solely with general-purpose transformer layers, could theoretically adopt Mamba in place of its Transformer backbone to reduce computational complexity and improve efficiency.

This question stems from my exploration of joint protein modeling. I am not requesting an empirical performance comparison between Mamba and Transformers within SimpleFold. Instead, I am seeking a theoretical assessment of feasibility and related insights. My broader research focuses on multimodal large language models for downstream protein tasks, which prompted my interest in SimpleFold’s implementation of Flow Matching.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions