Search papers, labs, and topics across Lattice.
1
0
3
Draft-OPD accelerates inference by over 5x while improving speculative decoding accuracy, transforming how draft models learn from target feedback.
