Search papers, labs, and topics across Lattice.
1
0
3
2
Forget generic pre-training: Speculative decoding gets a serious speed boost when your draft model is a specialist trained on data matching the target task.
