Skip to content

Metal installation documentation #1968

@devtobi

Description

@devtobi

I tried setting up llama-cpp-python in the current version 0.3.7 on my MacBook M4 Pro.
In the first step I only installed via pip install llama-cpp-python --no-cache-dir without specifiying the environment variable for Metal backend support.
I set the n_gpu_layers to -1 to fully use the GPU.

The interesting thing is the GPU was used even without having to install the Metal backend support as stated in the current documentation. I double checked this with a fresh start and explicitely setting the CMAKE_ARGS env variable and did not see and difference in terms of performance or GPU usage.

This is pretty handy, because when not using pip for dependency management (e.g. poetry) passing the environment variable did not work on my side.

Maybe the documentation should be updated to state that the env arguments are no longer required? This would also reflect the documentation in https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md#metal-build where it states that: "On MacOS, Metal is enabled by default"

This would also mean, that custom pre-built wheels are no longer required as well.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions