How did you fine tune your chatbot?

Posted by amohajerani1 on Sunday July 09, 2023 @03:21PM
from the self-post dept.

I want to fine tune a chatbot for therapy use. OpenAI has only a couple of tunable models, which are all completion models and developed in 2019.

It is depressing to be stuck with a 2019 model.

I am curious how others have approached this problem. How did you fine tune your chatbot?

( 3 comments | 3 points | on HN )

Re:How did you fine tune your chatbot? by edisonsama on Sunday July 09, 2023 @03:47PM (#36658146)

Hey,

I heard that lightchat.co is working on a highly personal chatbot that can also be used for work. So you could probably use that when it becomes available if you are willing to wait.

It's live but it's not personalized yet

[ Parent | on HN ]

Re:How did you fine tune your chatbot? by brucethemoose2 on Sunday July 09, 2023 @04:01PM (#36658301)

Try out Samantha, a llama 33b finetune, which may already be what you want: https://erichartford.com/meet-samantha

Run this one on exLLaMA\tui if you happen to have a 3090/4090 to run it on: https://huggingface.co/TheBloke/Samantha-33B-SuperHOT-8K-GPT...

And otherwise this on any decent PC: https://huggingface.co/TheBloke/Samantha-33B-SuperHOT-8K-GGM...

That would be a good starting point for a finetune if its not quite right.

[ Parent | on HN ]

Re:How did you fine tune your chatbot? by muzani on Monday July 10, 2023 @01:01AM (#36662308)

The "2019 model" is still better than most of the AI models out there, including Bard. Also they're releasing a new one very soon, but it'll likely be more restricted. The downside is the comparatively high prices and therapy is on the border of violating its usage policy (diagnosing people with mental health): https://openai.com/policies/usage-policies

And yeah, I fine tuned mine on curie. It has some bugs, but does what I expected it to do (help with writing dialogue).

[ Parent | on HN ]