Download path

[sarperdag]

Amazing library!

I'd like to check if a model is downloaded or not before starting a download, however, I can't seem to find a way to check this. How can I see if a model is fully downloaded and ready to use before calling LLMEvaluator's load method?

[davidkoski]

You would need to use https://github.com/huggingface/swift-transformers Hub API to check that. That said, the libraries here make use of that.

ModelFactory returns a container with a fully downloaded model:

• https://github.com/ml-explore/mlx-swift-examples/blob/main/Libraries/MLXLMCommon/ModelFactory.swift#L76

// e.g. VLMModelFactory.shared
let modelFactory: ModelFactory

// e.g. MLXVLM.ModelRegistry.paligemma3bMix4488bit
let modelConfiguration: ModelConfiguration

let container = try await modelFactory.loadContainer(configuration: modelConfiguration)

The is the code used to do the download -- you could call this directly if you were not using the code above:

• https://github.com/ml-explore/mlx-swift-examples/blob/main/Libraries/MLXLMCommon/Load.swift#L21

Ultimately that calls the Hub API:

            let repo = Hub.Repo(id: id)
            let modelFiles = ["*.safetensors", "*.json"]
            return try await hub.snapshot(
                from: repo, matching: modelFiles, progressHandler: progressHandler)

See if that works for you!

[yliuiuc1990]

How do I delete these downloaded models?

[davidkoski]

If you are doing the command line tools they are in ~/Documents/huggingface/models. For the applications they are in something named after the bundle id, e.g. ~/Library/Containers/mlx.LLMEval. You can delete those directories entirely or selectively the files in them.

[yliuiuc1990]

Got it. Was able to get the URL in Debug mode and delete the unused model files. Thank you @davidkoski