Instructions to use devin-lai/MiniMax-H3-Image-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use devin-lai/MiniMax-H3-Image-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download devin-lai/MiniMax-H3-Image-mlx --local-dir MiniMax-H3-Image-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Prepare inference-only v0.5.0 runtime cleanup
Prepare the current Hub runtime branch for the inference-only GitHub release v0.5.0.
- Replace the runtime, CLI, tests and docs with the audited GitHub release.
- Remove conversion, packing encoders, online quantization, Core ML export experiments and private model-publishing/preparation tooling.
- Replace the development report trees with the curated benchmark data and validation receipts already published on GitHub.
- Retain Hub card metadata, licensing, component metadata and the immutable v0.4.1 model catalog; refresh files.json for the resulting tree.
Validation: 61 local tests and portable GitHub CI on Python 3.10/3.12 pass for the source release. Installed-wheel BF16, packed12 and Q8 reproduce their prior PNGs and all 16 trajectory tensors. Q4 has loader tests, pinned remote manifest checks and separately labeled historical image evidence. The staged runtime bytes match the GitHub release, and the inference-only source audit passes.
This pull request is prepared for review and has not been merged. It changes the runtime repository only. Weight/shared/gallery repositories, gating, existing tags and model revisions are unchanged. Removing files from main does not remove them from old public revisions or prior copies. No history rewrite is included.
Source release: https://github.com/devin-lai/MiniMax-H3-Image-MLX/releases/tag/v0.5.0