[FEA] Enable reduced precision in IVF-flat, IVF-pq, and CAGRA indexes - Githubissues

rapidsai / raft

RAFT contains fundamental widely-used algorithms and primitives for machine learning and information retrieval. The algorithms are CUDA-accelerated and form building blocks for more easily writing high performance applications.

https://docs.rapids.ai/api/raft/stable/

Apache License 2.0

767 stars 194 forks source link

[FEA] Enable reduced precision in IVF-flat, IVF-pq, and CAGRA indexes #1675

Open cjnolet opened 1 year ago

cjnolet commented 1 year ago

In addition to the storage savings, this change also improves the data transfer rate and allows tensor cores to be used.

It can also provide a savings during index construction. For the IVF methods, k-means will need to be updated to support this as well.

[ ] CAGRA (https://github.com/rapidsai/raft/issues/1890)
[ ] Balanced K-means (https://github.com/rapidsai/raft/issues/1892)
[ ] IVF-PQ (https://github.com/rapidsai/raft/issues/1675)
[ ] Brute-force (https://github.com/rapidsai/raft/issues/1891)

cjnolet commented 9 months ago

Patrtially addressed in #2077