Timeline for How do the linear + softmax layers give out word probabilities in Transformer network?
Current License: CC BY-SA 4.0
4 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 8, 2021 at 19:20 | vote | accept | k-c | ||
| Jul 8, 2021 at 18:37 | comment | added | noe | At training time, everything is predicted in one pass. At inference time, the predictions are generated token by token. You can check this answer for details on this. | |
| Jul 8, 2021 at 18:34 | comment | added | k-c | So the decoding happens in just one pass? What I have seen is that the encoding happens in one go, then we sequentially generate the output sentence one word at a time, each time giving the sentence generated up to that point as the input. | |
| Jul 8, 2021 at 18:05 | history | answered | noe | CC BY-SA 4.0 |