Design a video vision-language model end to end

Design a model that understands video and answers questions about it in language: data and its labels, frame sampling and the video encoder, how visual tokens reach the language model, training stages, evaluation, and serving costs for long videos. Only the title of this report was public.