Self-Hosting Embeddings for DiscourseAI

We are storing embeddings using half precision (half storage space) and using binary quantization for indexes (32x smaller) by default as of a few weeks ago, so just updating your site to latest should give you ample disk usage reduction.

2 Likes