Hi I am an ecommerce company, i have 50k skus, i want to build a chat bot to whom customers can talk and get price or my skus.
I tried finetuning 3.5 turbo with conversation where user asking for price of specific skus.
But fine tuned model results mix of training data and from its own knowledge.
I want that model should use only my data for response.
Have you got any success in finetuning when data contain numbers ( in my case Price of skus)?
Hi and welcome to the developer forum!
Fine tuning is the wrong solution to your use case, this seems like a case for Embeddings, you can find the documentation here : OpenAI Platform
They will allow you to store your 50k product SKU’s in a vector database and allow users to semantically search them for relevant products, you can then put that data into a GPT model and chat with the customer about them.
Thanks Faxobilo, I will try that.
Search for RAG systems. With 50k it should be easy the retrieve the correct sku and inset it into the GPT context .