0

I've been trying to build a simple QnA bot using GPT-3 embeddings by following this specific guide.

Building a Custom GPT-3 Q&A Bot Using Embeddings

However, when doing so, I've been facing one huge issue when calculating vector similarity. Firstly, the issue have always been the two np.array() arrays, doc_embedding and query_embedding not having the same size. I've also tried the following two fixes:

  1. reducing the size of the larger array to the same size as the smaller array
  2. padding the smaller array to match the size of the larger array

Even though those two fixes stated above no longer raises an error, the given results are unsatisfactory. It isn't matching to the document sections as shown in the tutorial. It's matching to these following sections instead.

The selected document sections after calculating vector similarity

jonrsharpe
  • 115,751
  • 26
  • 228
  • 437
Min UD
  • 99
  • 1
  • 1
  • 12
  • 1
    yup, i was using an older embedding model which doesnt generate 1536 output dimensions. changing my model to the new embedding model works now! – Min UD May 09 '23 at 01:48

0 Answers0