How do i pre load the model with information

Assume i have a json file with information like {ItemID:ID,Price:price,…} how can i use embeddings(?) to be able to ask a question like: What is the price of item x?

Welcome to the community @arau.j

If the whole thing is structured uniformly with JSON, then in this case you don’t need embeddings.

Simply describe the structure of JSON in the prompt and then translate natural language to JSON.

Hi, thanks for the reply! I’m sorry I don’t think I expressed my self clearly. The goal is to import a json with > 1000 items with prices, descriptions etc. then also import another json with for example orders (multiple thousand rows) and be able to ask questions like „what is the price of item x“ or „how many times was item y sold last year“ etc. I don’t think that your solution would work here right? Because it’s far beyond the allowed prompt length.

1 Like

You could use my Vectra Project to create a local Vector DB you can use semantic search over to include relevant snippets into your prompt. There’s also a python version of Vectra.