Fine tuning a model for customer service for our specific app

The second example will probably work better. I haven’t tried anything in the first format you gave

I have been looking for this information everywhere, finally found it here. Thank you!

I do have one more question: if I need to train the model on multiple full conversations, where there is a back and forth in between the user and the agent, how do I group the individual conversations given the above format?

To explain a little better, in the examples shown so far, each prompt > completion works as a standalone. If possible, I would like to be more specific with a series of full conversations, in which the prompt > completion depends on the overall context of the individual discussion.

Hi @LeoBit,

We have a similar use case and are considering following the approach mentioned in this documentation

Please let me know if this information is helpful or if you are already exploring an alternative approach.

can you help me with this because i want to train my data with openapi fine tuning after this is used by fine_tuned_id for my customer support chatbot but when i try to train my data with open api key (gpt-3.5) when i ask questions after training i get reasked questions from my own dataset so what a shame This fine-tuning model does not respond well and mostly gives irrelevant information or general information. How can I solve this problem? Do you have any reference code or source code to solve this problem? My data set consists of 173 questions and answers in a csv file in the sample form below “{”“messages”“: [{”“role”“: ““system””, ““content””:”“A2Asistan, which aims to help people. You are an extremely friendly customer chatbot who loves and you are not satisfied unless the customer is completely satisfied.”“},{”“role”“: ““user””, ““content””:”“How can we install Jivo Chat program on our website? .”“},{”“role”“: ““assistant””, ““content””:”“The JavaScript code you will receive from JivoChat will be added to a2admin > Site Management > Settings > Long Texts > tag. Just add it to the : field.”“}]}”