原书出处
Designing Large Language Model Applications
margin sampling, as proposed by Ramirez et al. In the margin sampling method, we generate the first token and use the difference in the probability of the most probable token and the second most probable token as the margin. The assumption is that the higher the margin, the more confident the model. If the margin is below a certain threshold, then the input is sent
原书定位详情
来源:a19c7933-4eda-4176-aebd-30fe7e1ad292
正文位置:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch13.html
EPUB SHA-256:1b4d2347fec2f949fcddfbaf4fad597a1ae20887a096cfae64c59a619b4781bf