This Waifu Does Not Exist: http://www.thiswaifudoesnotexist.net/index.html (u mad?)pic.twitter.com/slt7dsYgXT
Показване на тази нишка
-
What prompts were used for the summaries? Was it literally just "Do Electrical Neurons Dream Of Anime Shows?" ?
No, I used a pretty long one to try to stuff a lot of keywords in for GPT-2 to use. Playing around, it seemed like the more keywords I added, the more interesting the outputs tended to be. So I wound up with this:https://pastebin.com/Y7Pb52iM
-
-
I noticed your gpt2 public model also seems to have trouble with actually ending on the end of text command. I wonder what is up with that, like half my gens have it when I tried it out.
-
I don't think the sampling code actually has any support for ending when the '<endoftext>' string gets generated, so it just keeps going and forcing more characters to be generated. I've started simply stripping them out with sed.
-
Well it decides to end at some point so... not sure? The text that follows is usually completely unrelated to the sample text.
-
I mean the Python code running the model. The model has no way to 'decide' to stop if the Python chooses to ignore it when it generates '<endoftext>'.
Край на разговора
Нов разговор -
Изглежда зареждането отнема известно време.
Twitter може да е претоварен или да изпитва моментно затруднение. Опитай отново или виж Twitter Status за повече информация.