[Install issue]: Supporting Library , Datatypes , Querytype and Index types

chroma-core / chroma

the AI-native open-source embedding database

https://www.trychroma.com/

Apache License 2.0

13.75k stars 1.16k forks source link

[Install issue]: Supporting Library , Datatypes , Querytype and Index types #1311

Open pauldintel opened 8 months ago

pauldintel commented 8 months ago

What happened?

hi Team , Like other vector DB I am looking what Libraries and Index type from below is supported by ChromaDB

HNSWlib DiskANN / Vamana ScaNN FAISS-IVFP NGT

Also are all these datatypes supported ? FLOAT INT8 , UINT8 FLOAT16 BFLOAT16

And does ChromaDB has support of Intel AVX instructions like AVX512 ?

What are the query types from below it supports ?

Search Filter + Search

thanks

Versions

Latest Chroma version

Relevant log output

NA

HammadB commented 8 months ago

Currently only HNSW indices are supported.

We only support full-width floats.

We do support AVX512 inside the hnsw index.

We do support filter+search https://docs.trychroma.com/usage-guide#using-where-filters

pauldintel commented 8 months ago

@HammadB thanks for the update . full width floats means FLOATT16 or 32 only ?

pauldintel commented 8 months ago

do you support below ? Generate, Store and Seach/Index Embedding hybrid (Semantic, Multi-modal, Gen AI) search

tazarov commented 8 months ago

@pauldintel, we support 32bit floats. Regarding the embeddings:

Generate - yes (via Embedding Functions like OpenAI, HF, Cohere and a default Mini
Store - yes (custom binary for vectors + sqlite for metadata)
Search/Index - yes, as @HammadB, hnsw lib for now

For search, as long as you can turn it into a vector, you can store it and search it. Semantic - via Embedding Functions, multi-modal - coming up soon