NVIDIA / NVIDIA/cuvs

[FEA] Out of Core Batched Search for IVF-Flat

Open
#1,292 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

feature request
Dominant language
Cuda
Stars
854
Forks
236
Avg merge
3d 3h
Merged PRs (30d)
62

Description

Reasoning: IVF-Flat can be an out of core graph building algorithm for CAGRA by just loading nprobe lists on to the device. This can work effectively for CAGRA graph building because there can be some smart batching methods to prolong resetting the loaded nprobe lists (for example by searching neighbors of all the vectors with the same coarse assignments as a single batch).
Something to ponder: To take this one step further, IVF-Flat can be a special case of brute force with all-neighbors.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No files, tests, or entry points are named. Start by locating the IVF-Flat and CAGRA graph-building implementations, then clarify the batching and nprobe-list loading design; done should mean an agreed implementation and validation path for out-of-core graph building.

Written by the indexing model from the issue text.

Assessment

Tech stack
search
Domain
machine-learning, search
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.