How can I fine-tune a GPT model to accurately handle rephrased or varied questions for my e-commerce chatbot?

API
,
1

I have an e-commerce website and want to integrate a custom chatbot. I’m using the OpenAI API with PHP code to communicate with a GPT model. However, when I fine-tuned the model, it only responds accurately when asked the exact questions from the training data. If the question is rephrased or asked in a slightly different way, it returns irrelevant information instead of the correct data from my website.

How can I fine-tune the model so it can handle varied or rephrased questions?

2

as far as I see, there are two options:

  1. Include all the data on a single prompt or do a mixture of experts approach (multiple agents that trigger other agents) - side note: make sure to include the moderation endpoint on this

  2. create enough data, with many examples and possible questions, to properly fine tune the modal

please refer to this post for more information on what the community has talked about, it is a great way to fast track your understanding of fine tunning

1 Like