Cascade Reward Sampling for Efficient Decoding-Time Alignment

3 points| Garcia98 | 1 year ago |arxiv.org