Search papers, labs, and topics across Lattice.
Stanford University
1
0
3
Unlock linear-time video VLMs without accuracy loss: StateKV matches full self-attention while crushing sliding-window methods, all without finetuning.
