I am using similarity_top_k=10 to query vector index. But it takes long time to get final answer based best matches. Is there a way to stream responses after completing prompt for each selected passage like ChatGPT? If query API doesn't provide that, Is it possible to manually achieve it by selecting similarity_top_k=1 ten times ?

@Logan M @jerryjliu0 I am currently using GPTSimpleVectorIndex, but the index file huge and having issues with loading in Google app engine. I am thinking about switching to GPTWeaviateIndex, but not sure if it solve problem.

When I save GPTWeaviateIndex to file to load it back during querying, will data be still stored in the local file, so file size will still be huge?

Is there a way to create index once and query it without saving to local disk?

Is there an example for providing custom regex to extract keywords from the document?. I suppose I should use GPTSimpleKeywordTableIndex, but not sure how to custom regex to it.