Search papers, labs, and topics across Lattice.
1
0
3
0
Sparrow unlocks 2.8x faster inference for Video LLMs on long videos by cleverly offloading visual computation to the target model using text-anchored attention and semantic-rich intermediate states.
