Skip to content

Multi-arch support for pre-built cpu wheel #1506

Open
@abetlen

Description

@abetlen

Currently pre-built cpu wheels are compiled for the lowest common denominator architecture. If we can compile multiple versions of the llama.cpp library with different accelerations compiled in each we can bundle all of them into the pre-built wheels and dynamically choose one based on the host cpu.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions