Search papers, labs, and topics across Lattice.
Affiliation:
2
0
4
LLM inference gets a 2x speed boost without training, thanks to a clever technique that merges retrieval with logit-based speculation.
