patelprateek

Distributed build of hnsw and merge of hnsw graphs

2

@yurymalkov : I have multiple indexers producing many sharded hnsw indexes , but servers perhaps can handle multiple such shards . I read in the paper that a distributed implementation...

Filtering based on expressions

5

New user here so please bear with me some naive questions. Is it possible to have some user code , math expression based filtering while doing graph search (hnsw) I...

cardinality estimation for filters and queries

1

new to vespa engine. I am trying to use this for some internal dataset Wanted to get better understanding of some internal details to understand latency and compute resource usage....

Thanks for this amazing work. Can you please provide some details on current design for query planner when we want to do vector search with some filtering constraints, i observed...

evaluating and extending hnsw for SL2G

3

came across this interesting paper : http://research.baidu.com/Public/uploads/5e88947b04a4e.pdf IIUC the method proposes to use l2 for building hnsw graphs and searching using some NN function. Curious about any such experiments or...

quantizer documentation

1

Thanks for the amazing work. I am reading the documentation and it says : 8-bit quantization: Quantile, Linear, and Dynamic quantization In the code i also observe block wise quantization....

auto sharding strategy for theta sketch

3

I was going through the pr : https://github.com/apache/pinot/pull/5316 Can you please point me to how or where is this implemented. How do we define high cardinality threshold I am running...

Support for 16 bit quantization

2

Are there any plans for supporting 16 bit floats (fp16 and Bfloat16) quantization for embeddings. I would assume it would be an easier choice that doesn't separately need to train...

design choice and benchmarks

1

I observe the usage of lmdb , sqlite , pq codec for hnsw can you please provide some rationale on design choices here for example : rocksdb instead of lmdb...