Search papers, labs, and topics across Lattice.
University of Maryland
1
0
3
LLMs can slash multi-hop retrieval latency by 40% using SpecHop, a framework that speculatively executes multiple reasoning paths and rolls back incorrect ones.
