[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/sci/ - Science & Math


View post   

File: 81 KB, 1024x989, 1660335578736140.jpg [View same] [iqdb] [saucenao] [google]
14883109 No.14883109 [Reply] [Original]

can someone explain or link article/tutorial about how generation functions work with transformers (NLP)?
i understand the architecture, how it works etc but i have no fucking clue what the parameters to the generation functions are, what beam search is, why for the same input different outputs can be generate etc and it's completely ommited in everything i come across
if i had to guess then since logits go into the classifier head that outputs basically probabilities of every word occuring next, then beam search/etc are ways to choose words from that probability vector?

>> No.14883238

ok sorry im retarded, found it now
kill the thread and my embarrassment