1 paper
Prasann Singhal, Jiacheng Xu, Xi Ye +1
Standard decoding approaches for conditional text generation tasks typically search for an output hypothesis with high model probability, but this may not yield the best hypothesis…