I’ve been trying to use fine tuned models with the assistants in the playground. Whilst the models can be selected they produce a generic error when I try to use them.
I’ve tried with a GPT-3.5 Turbo fine tuned model (release notes from April '24 say this is supported with Assistants) and with a GPT-4o mini model (documentation is ambiguous). Neither works in the playground.
I’ve found older similar topics to this, but no concrete answer why this isn’t working and what is the expected behaviour.
I am experiencing a similar problem.
Today I finetuned a model but it is not loading in the playground, probably an error in the finetuned models endpoint.
Both models I have created are fully evaluated (according to the UI) and active. I can select them in the assistant configuration. Based on the documentation and the fact they are selectable in the UI makes me think this is a bug.
I’m having very similar issues. Models were successfully fine tuned and I can use them directly but instantly fail on any prompts when used in an assistant.
Unfortunately not. I’ve resorted to using my training material within the assistant’s instructions instead, which is performing OK, but isn’t really cost effective given the amount of tokens it uses.