Search papers, labs, and topics across Lattice.
Zhejiang University
1
0
3
Smaller models get a bigger speed boost from Speculative Decoding on software engineering tasks, challenging the assumption that larger models always benefit more from inference acceleration techniques.
