#Can I create embeddings/finetune chatGPT3.5Turbo and use that with the chat api? How?
22 messages · Page 1 of 1 (latest)
I've got the system prompt the way I want and it works great, but I can't include all of the data (thousands of records) in the system as it overloads the token limit.
You cannot fine-tune turbo. You also need to use embeddings instead
Thanks can you point to any resource on how to use embeddings with turbo?
or is it the same, I just substitute turbo in for what I'm already doing with davinci?
The main goal is to add internal documents to ChatGPT and have it return those in its responses through the chat api (not completions).
Can you point me to any resources? I’ve been successful with completions but my goo-fu must be failing me with the chat api.
look in the openai community forums
Is that not this?
langchain also has plugins for that
no, this is discord
this is the community forum https://community.openai.com/
Thanks I’ll check both of those out.
hi, sorry im still new with finetune, i couldnt get it to work, i did a simple prompt and completion with jsonl and created finetune with success via davinci model, but when i use the model finetuned, it never give me the same completion as my prompt to completion, for example prompt: what is xyz? completion: xyz is a product abc brand. finetune prepared without any error as i followed the ending suffix for prompt and completion with ending \n\n. been figuring for 1week now. but @amber ruin i would love to hear if you found your solution for chat, it will be even better
Looks like you're doing Q/A, you should use embeddings