Built with @LysandreJik, @Thom_Wolf and the whole @huggingface team during @naaclhlt last week. (hat/tip to the @NeuralGen workshop)
Show this thread
At NAACL last week we built a new side project, Write With Transformer.
It lets you trigger GPT-2 completions multiple times, in a Google Doc-like interface.
It's like having a unicorn friend that completes your thoughts cc @gdb @AlecRad
Try it: https://transformer.huggingface.co/ pic.twitter.com/betwsNe3tR
-
-
-
It was inspired by
@AdamDanielKing's awesome http://talktotransformer.com and more generally by the lineage of demos around language modeling such as the great@robinsloan experiment Writing with the machine.Show this thread -
Ok so
@LysandreJik switched to a 8xV100 instance so the demo should be more BLAZING FAST than ever.
#gpuRulesShow this thread -
One of the perks of building this app was that I had to make a vectorized version of a unicorn, which I wanted to do for a long time :) Hi :)pic.twitter.com/xJxMcdzh41
Show this thread
End of conversation
New conversation -
-
-
I've been waiting for someone to do this!
I would pay for pro version: integrates GPT-2 with your OS keyboard (so it works in any app), and offers choice of models or even upload custom model (possible very narrow domains, like a Python ML code model) -
That is a NEAT idea. We might take you up on that
End of conversation
New conversation -
-
-
This is lovely!
-
Thanks Jeremy! Already 41k completions served according to gunicorn
End of conversation
New conversation -
-
-
What a sneaky app!
Meta-beginning + a slightly modified quote from Dune = guilt-free indulgence
pic.twitter.com/x76oK88grP
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
We're calling our novel "WIND OF THE UNKNOWN" and are willing to split the royaltiespic.twitter.com/yaWA12xuQL
- 1 more reply
New conversation -
-
-
This is really interesting. Can you explain how the sliders work?
-
Temperature sharpens/flattens the next word probability distribution. Low temperature emphasizes more-probable words (common ending), high temperature: more diversity. Top-p controls the size of the pool of words: similar effect (high => ++ diverse). Code:https://gist.github.com/thomwolf/1a5a29f6962089e871b94cbd09daf317 …
-
Thank you!
End of conversation
New conversation -
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Do you know if the writings are predictive? Would love to see that for real!
#justice#chocolatinepic.twitter.com/w6gRWHEYIy
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Great job!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Waouh
really amazing..I love deep learning and NPLThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Any way to load a fine tuned model? Do you think you’ll add this to your github?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.