Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
1
0
2
Speculative decoding doesn't need to strictly match the LLM's distribution – relaxing this constraint with Cactus yields faster decoding without sacrificing quality.