Skip to content

[New Model]: Cohere2 (Command R7B) #11181

Closed
@fizzAI

Description

The model to consider.

https://huggingface.co/CohereForAI/c4ai-command-r7b-12-2024

The closest model vllm already supports.

Likely either the original Cohere (for. obvious reasons) or Gemma2 (as it also has a funky SWA architecture)

What's your difficulty of supporting the model you want?

It uses SWA, but this can likely be ditched to get MVP inference working ala how gemma 2 was done
For some reason every 4th layer uses global attention without positional embeddings? Not sure how or why that one works tbh

Before submitting a new issue...

  • Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

Metadata

Assignees

No one assigned

    Labels

    new modelRequests to new models

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions