Agreed, I've run sentence-transformers/all-MiniLM-L6-v2 locally on CPU for a similar task, and it was approx X2 faster than calling the OpenAI embedding API, not to mention free.