We released a new 7B VLM and large-scale dataset for video understanding. Arxiv. Dataset. (code release soon, in the hands of corporate overlords)