PythonPro
kiwifarms.net
- Joined
- Aug 19, 2022
Hello, I trained a gpt2 algorithm with most of Chris Chan’s written and spoken works, including but not limited to
- All the Sonichu comics
- His entire twitter
- Most of his prison letters
- Most of YouTube videos, helpfully transcribed by members of the Cwiki
- Selected letters to his sweethearts/trolls
After debuting it on Null’s show I am opening it to the public to see if they can make it say anything interesting.
I included a link to the model, along with the text I used to train it in case you want to run it on your own hardware. All in all the text data is about 2 megs, probably much less than the bare minimum for this sort of thing. I didn’t clean the data very well hence the bizarre responses and artifacts, I filter certain ones out on the website. I ran the algorithm for thirty epochs, for a combined total of several days (I couldn’t get it to use my GPU). If you have any questions I’ll try to answer them.
https://anonfiles.com/pbQcq5A2y6/CWCbotData_zip
Keep in mind that I am running it on the shittiest Google VPC and will probably take it down when the free trial runs out.
The queue size is currently at 5, any more and the last guy is guaranteed to time-out. Be patient. It is working. It just takes over a minute to generate just 250 tokens.
If you get any interesting responses, feel free to post them in the thread.
The link is:
http://www.cwcbot.xyz/input/
- All the Sonichu comics
- His entire twitter
- Most of his prison letters
- Most of YouTube videos, helpfully transcribed by members of the Cwiki
- Selected letters to his sweethearts/trolls
After debuting it on Null’s show I am opening it to the public to see if they can make it say anything interesting.
I included a link to the model, along with the text I used to train it in case you want to run it on your own hardware. All in all the text data is about 2 megs, probably much less than the bare minimum for this sort of thing. I didn’t clean the data very well hence the bizarre responses and artifacts, I filter certain ones out on the website. I ran the algorithm for thirty epochs, for a combined total of several days (I couldn’t get it to use my GPU). If you have any questions I’ll try to answer them.
https://anonfiles.com/pbQcq5A2y6/CWCbotData_zip
Keep in mind that I am running it on the shittiest Google VPC and will probably take it down when the free trial runs out.
The queue size is currently at 5, any more and the last guy is guaranteed to time-out. Be patient. It is working. It just takes over a minute to generate just 250 tokens.
If you get any interesting responses, feel free to post them in the thread.
The link is:
http://www.cwcbot.xyz/input/