Search papers, labs, and topics across Lattice.
2
0
3
4
Visual updates in DeltaV cut token generation by over half while boosting reasoning accuracy, challenging the need for full-image outputs in multimodal models.
VideoLLMs can now watch and think *simultaneously*, achieving 15x faster response times and improved accuracy on video understanding tasks.
