Does Milvus supports partial loading of collection in memory to perform similarity search? I mean, based on the input vector, will it be able to identify and auto-load clusters of vectors which most likely are similar?
If no, is there any vectordb (like faiss, nmslib etc) which supports partial loading of indexes in memory?
You're probably looking for indexes like DiskANN - it holds ~10% of the vectors in memory and the rest on disk
I believe IVF Flat also holds only the centroids and then does a flat search on the vectors within the closest centroids.