TensorLearn
Back to Course
LLM Engineering: Transformers & RAG
Module 9 of 12

9. The Math of Embeddings

1. High Dimensional Space

You cannot visualize 1536 dimensions (OpenAI Ada-002). But math works the same. $$ A cdot B = |A||B| cos( heta) $$

2. Search

Finding the "closest" document is just finding the vector with the smallest angle. We use Approximate Nearest Neighbors (ANN) indices like FAISS to do this in milliseconds for billions of vectors.

Mark as Completed

TensorLearn - AI Engineering for Professionals