docs: link the companion motion-corpus dataset repo

docs: document the published v8.0 dataset + corpus-free reproduction

data: publish the 150-clip curated library v8.0 was trained on

docs: refer to the validation clip generically as a reference walk

docs: v8.0 model card — template-taught, leg-chain fix, 30 actions (#837)

motion-library-v2: user-curated 41-clip set from the #954-fixed extraction (root-only FBX space-change, facing-aware references, refRoll) — Woman clips now correct; Fruit Wave removed

Rollback motion-library-v2 to July extractions: the fresh re-extraction was poisoned by an FBX importer arm-chain pitch regression (punch/run arms wide/overhead on retarget)

Upload motion-library-v2.json with huggingface_hub

Upload motion-library-v2.json with huggingface_hub

Upload motion-library-v2.json with huggingface_hub

Upload motion-library-v2.json with huggingface_hub

V2 attribution (#838)

motion library V2 (schema v4): 36 user-curated clips (#838)

sync: refresh ATTRIBUTION (#838)

sync: vertical descent + spine guard + curation (#838), 46 clips

Upload motion-library.json with huggingface_hub

Upload README.md with huggingface_hub

Update card for v4 architecture + corrected 6D

v4 vocab (frame=world, fps=30, 13 actions)

v4 corrected-6D model — column-packed rotations, world-frame, 30fps (#411)

Experimental from-scratch text-to-motion model (#411): CMU+Quaternius CC0, ONNX

initial commit