Mantain context and conversations with text-completions

[sterile galleon]

Is it possible to keep the conversations stored in OpenAI so I don't have to send all the conversation content every API request? Doing this every API call ends up consuming a lot of tokens and significantly increases the overral cost of the API. Currently I'm storing the conversations and contexts in the backend and resending the entire context+conversation history with each request.

[tame latch]

Currently there’s no way to store context upstream, so it has to be handled locally for now

I also have a project which does the same thing (storing and resending the entire context + convo)

[thorn slate]

I thought there is the possibility to rain you own model via the API?

[tame latch]

Yes, but that doesn't store context

[thorn slate]

https://beta.openai.com/docs/guides/fine-tuning -> there is file upload also possible?

[thorn slate]

I'm curious because I read about it, I use the API and I'm planning to use fine tuning in a project, but have not tried it.

[thorn slate]

I'm curious because I read about it, I use the API and I'm planning to use fine tuning in a project, but I have not (yet) tried it. So it is supported but not working?

[thorn slate]

or the usage of files: https://beta.openai.com/docs/api-reference/files -> isn't this the possibility to store context also?

[tame latch]

Fine tuning is with prompt <> completion pairs

It's possible to make it act as context but difficult, you need a LOT of training data.

[tame latch]

The files that it's talking about here are the jsonl files that are used to train the model in fine-tuning

[thorn slate]

Have you done it practically?

[tame latch]

I have not fine tuned myself, no, but I've used embeddings to store context