Search papers, labs, and topics across Lattice.
Beihang University
1
0
3
CrossPool achieves a staggering 10.4x reduction in tail latency for bursty long-context requests by decoupling model weights from KV-cache in GPU memory.
