Instructions to use lightx2v/microduck with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Microduck
How to use lightx2v/microduck with Microduck:
# Replace SLOT with the slot specified in the model card (walk, stand, sitstand, ground_pick, kick_left, kick_right, roulade). sudo robotctl policy load SLOT lightx2v/microduck
- Notebooks
- Google Colab
- Kaggle
|
Download NOTICE.md from lightx2v/microduck: direct link, hf CLI and curl.
- Browser
- Download file 2.45 kB
-
https://huggingface.co/lightx2v/microduck/resolve/main/NOTICE.md
- Command line
-
hf download hf://lightx2v/microduck/NOTICE.md
-
curl -L -o NOTICE.md https://huggingface.co/lightx2v/microduck/resolve/main/NOTICE.md
2.45 kB
Artifact provenance and third-party notices
This is the inference-file release for https://github.com/fuheaven/LightX2V-microduck.
Project-trained navigation checkpoint
- Original run:
clear-room-realtime-control-pilot-v1, step 250. - Exact model SHA256:
8ea7793290f0a22bff56ec4e44168232564ec5d57a9e21515a8a8c4a989114ea. - The original normalization and checkpoint receipt are preserved byte for byte.
- A separate license for these project-specific weights has not been designated by the owner. The presence of third-party license texts below is not a blanket license assignment to these weights.
- Hardware status:
hardware_unvalidated. Published numbers are simulation development evidence, not independent full acceptance.
Wan inference components
- Source: https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B/tree/921dbaf3f1674a56f47e83fb80a34bac8a8f203e
- Files: the seven files under
models/Wan2.2-TI2V-5B/inmodel-files.lock.json. - License identified by the upstream model card: Apache-2.0. A copy is provided in
licenses/Apache-2.0.txt; the original model card is retained inlicenses/Wan2.2-UPSTREAM-README.md. - UMT5 tokenizer and text-encoder components retain their upstream attribution to the UMT5 project: https://huggingface.co/google/umt5-xxl.
- All mirrored files retain their original bytes and SHA256. The original diffusion-model shards are not redistributed in this bundle.
Implementation and research attribution
- LightX2V: https://github.com/ModelTC/LightX2V, Apache-2.0.
- FastWAM: https://github.com/yuantianyuan01/FastWAM, MIT source license retained as
licenses/FastWAM-MIT.txt. - The Python inference implementation, Docker build, simulator, tests, and evidence are distributed by the linked code repository. This model repository does not ship training optimizer states or an executable Python pickle loader beyond the original model serialization.
Separately downloaded simulator dependencies
- MicroDuck RL body: https://github.com/pollen-robotics/microduck_rl at
2b581c641406a48346e696212930ea881c222c52. - Official policies: https://huggingface.co/pollen-robotics/microduck-policies at
088524a64e2557dc453256b6071dbb9d23888802. - Robot 3D assets have separate CC BY-NC-SA terms stated upstream; this release does not replace them with the code license. They are not bundled here.
- The code repository's pinned fetch command obtains the original assets and policy documentation.