MiniMax H3 AI Pod inference runtimes
Verified ARM64 runtime archives for Thor T5000 (JetPack 7) and AGX Orin 64GB (JetPack 6). These contain inference code and dependencies, not model weights. Weights are downloaded separately from OpenVDN/vdn-minimax-h3 and Comfy-Org/MiniMax-H3. LPK 0.5.1 uses these unchanged runtime images.
| Target | Archive | Bytes | SHA256 |
|---|---|---|---|
| thor | minimax-h3-thor-0.4.6-runtime.tar |
14220806144 | 83fbd4b01b0f94265d33c51f0c70296091593e9bfaae23cdabbcc7fcfff26ca0 |
| orin-64g | minimax-h3-orin-0.5.0-runtime.tar |
9809733120 | f482b8b814071cffb4016bb09fea24e589d6333a0c523f98134e5e7869e5b0ce |
| thor-cache | minimax-h3-thor-0.4.6-warm-cache.tar.gz |
7452828 | 5b9611779a365080485fd746263e8b2bfc1c98af43b9297e81b97dfd3c55216f |
Verify SHA256, then run docker load -i <runtime.tar>. Verify the imported image identity against checksums.json. The warm-cache tar.gz is a separate Thor compiler-cache seed, not a Docker image. Orin builds its own cache.
Third-party software retains its original licenses and notices inside the images, including NVIDIA runtime terms. This repository does not relicense those dependencies. See NOTICE.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support