Search papers, labs, and topics across Lattice.
1
4
3
Speculative decoding for Llama just got 10% faster, thanks to production-scale optimizations that unlock new levels of inference efficiency.
