Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
Retaining the largest attention weights in KV-cache eviction is nearly optimal, challenging the perceived complexity of selection strategies.
