Instructions to use SceneWorks/sam3-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use SceneWorks/sam3-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir sam3-mlx SceneWorks/sam3-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
File size: 1,292 Bytes
fc87116 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 | ---
license: other
license_name: sam-license
license_link: LICENSE
tags:
- segmentation
- sam3
- mlx
- mirror
---
# SceneWorks/sam3-mlx
A **redistribution mirror** of [`facebook/sam3`](https://huggingface.co/facebook/sam3)
(Segment Anything 3) for download-on-first-use by the [SceneWorks](https://github.com/michaeltrefry) macOS worker.
The native-MLX (Rust `mlx-gen-sam3`) SAM3 segmenter loads the **stock checkpoint directly** — no
conversion — and drives text-concept Promptable Concept Segmentation ("person") for the
Replace-Person person-track pipeline (epic 4910). This mirror exists only so the app can fetch the
weights without the gated `facebook/sam3` access flow.
## Files
- `model.safetensors` — the SAM3 checkpoint (~0.9B params, F32), byte-identical to `facebook/sam3`.
- `tokenizer.json` — the CLIP-H text tokenizer used for the concept prompt.
- `LICENSE` — the Meta **SAM License** (included as required by the license; it governs use of the Materials).
## License
These weights are © Meta and distributed under the **SAM License** (see `LICENSE`). This mirror
redistributes them unmodified under that license. Refer to the original
[`facebook/sam3`](https://huggingface.co/facebook/sam3) model card for intended use, capabilities,
and limitations.
|