[binarymax]

Hey there - interesting solution!

A note on benchmarks with HNSW is that you need to optimize your ‘m’ and ‘ef_construction’ params at index time and your ef_search param at query time.

So you may not be getting an optimized result from DBs like Qdrant and weaviate.

[lqhl]

That's true and we have tried different configurations for systems that use HNSW. But for the ease of presentation, we only choose the configuration with the highest throughput at precision 98%.

Here is a figure in our open-source benchmark framework repo that shows other configurations that we have tested: https://github.com/myscale/vector-db-benchmark/blob/master/i...

[renaissancist]

we actually tuned `m` (e.g. 32), `ef_construct` and `ef_search` (e.g. 128-256) to be fairly large values for best performance, but not able to obtain better results. Overall HNSW consumes a lot of resources, so it's not as cost effective.