#Fine Tune and token limits
inland fern
You'll want to read up on token limits
4000 xd
sacred bobcat
Depending on the model used, requests can use up to 4097 tokens shared between prompt and completion. If your prompt is 4000 tokens, your completion can be 97 tokens at most.
1 token is approximately 4 characters (~0.75 words).
84,000 words ~ 112,000 tokens.
Depending on what you're trying to do a fine tune model may not be necessary
If it is then you'll need to break up the prompts to fit the token limit
proven pumice
>Depending on the model used, requests can use up to 4097 tokens shared between ...
I don't think this is accurate I think that's for general GPT3 prompting, For fine tuning, each prompt and completion can't exceed 2048 tokens including the separator
Fine tuning is what you want here, not the regular GPT3 usecase of sending a prompt with 4096 context
For your book/text, you can create 2048 token prompt/response pairs to fine tune the model with, but it will take some work to determine how to structure those prompt/response pairs
Currently, there's no way to just input a full text to train the model
sacred bobcat
thank you for the correction, Kaveen! That is really important to know.
proven pumice
thank you for the correction, Kaveen! That is really important to know.
No worries! 🙂
inland fern
I'm going to try smaller texts,