We just started using Semantic Search using the built-in models in our Staging cluster and ram usage has spiked. We have very little documents in this cluster is it the vector embeddings taking up this space? Is the model downloaded onto the cluster?