EmpathicRobotics/omnivideo-100k-final

omnivideo-100k-final QA-restructure + instruction-diversity augment (v5) release of the OmniVideo-100K (window=8) training text -- 5,214 rows. The video-token half of every record (<seed2_N>/<cosmos_N>/<agent>/<listen>/<caption>/ <speech> chunks, plus the leading "Continue this video." summary turn) is byte-identical to v3/v4 -- Step A never reads the QA files, so none of this pass touches video/pose tokenization. Only the trailing QA content changed. What changed… See the full description on the dataset page: https://huggingface.co/datasets/EmpathicRobotics/omnivideo-100k-final.

<iframe src="https://silytics.si/embed/item/697fc5c3-ecdd-4e28-893d-fe2d931fb606" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.