A non-parametric yet learnable router selects informative history chunks to calculate attention. The routing is implicitly differentiable and optimized end-to-end from large-scale long video data.
For 8-shot, minute-long videos (~180k tokens), MoC prunes ≈ 85 % of token pairs and cuts FLOPs by 7× while delivering minute-long context coherence.
Some prompts are from Twitter artworks.
On 8-second, 320×192 clips (~6.5k tokens), MoC routing prunes ≈ 83 % of token pairs, while maintaining or even improving video quality.
Side-by-side comparisons between dense attention and our Mixture of Contexts.