Skip to content

Request: Public 0.5B model weights for on-device / edge deployment #333

@saberrg

Description

@saberrg

The README describes Moondream 0.5B as "optimized as a distillation target for edge devices" for deployment on "mobile devices, Raspberry Pis, or embedded cameras." However, there doesn't appear to be a public download for the 0.5B weights, only the 2B model is available on HuggingFace (vikhyatk/moondream2). The config already exists in the repo (config_md05.json), and the native loading code in moondream/torch/weights.py supports it. What's missing is a .safetensors or .pt checkpoint file.

Could you make the 0.5B weights publicly available?

This would enable developers to convert and deploy Moondream on mobile platforms (e.g., CoreML for iOS, TFLite for Android) where the 2B model is too large to run alongside other on-device models. The 0.5B at 4-bit quantization would bring the model bundle under 100 MB, which is practical for App Store distribution.

If the 0.5B is only available through the Lens fine-tuning platform, it would be helpful to document that, right now it's unclear from the README and repo how developers are meant to access it.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    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