# NVIDIA/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.

Repository: https://github.com/NVIDIA/raft
Canonical: https://ross.abutalabs.com/products/nvidia-raft
Homepage: https://docs.rapids.ai/api/raft/stable/
Language: Cuda
License: Apache-2.0
License Family: permissive
Topics: anns, building-blocks, clustering, cuda, distance, gpu, information-retrieval, linear-algebra, llm, machine-learning, nearest-neighbors, primitives, random-sampling, solvers, sparse, statistics, vector-search, vector-similarity, vector-store, neighborhood-methods
Last push: 2026-08-31T22:16:18+00:00

## Health v2 (maintenance only)
Score: 94/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 100, release rhythm 84, longevity 100
- inputs: {"age_days": 2638, "days_push": 2, "days_rel": 28, "gap_med": 62, "n_releases_24m": 12}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1038, forks 248 (observed 2026-09-01T02:13:55.828783+00:00)

## What it is
NVIDIA RAFT (Reusable Accelerated Functions and Tools) is a C++/Python library of CUDA-accelerated algorithms and primitives for machine learning and information retrieval. It provides building blocks such as linear algebra, sparse and dense operations, solvers, statistics, and nearest-neighbor search that underpin the RAPIDS ecosystem.

## Use cases
- accelerate nearest neighbor search on gpu
- gpu-accelerated linear algebra primitives for ml
- build high performance cuda applications from reusable building blocks
- sparse matrix operations on gpu
- vector similarity search for llm applications
- multi-node multi-gpu algorithm development
- sampling and statistics computations on gpu

## When to choose
- you need CUDA-accelerated primitives like ANN search, distance computations, or linear algebra in C++ or Python
- you are building GPU applications within the RAPIDS ecosystem
- you want header-only C++ libraries to reduce build and maintenance burden
- you need multi-node multi-GPU communication abstractions for distributed algorithms

## When to avoid
- you need a complete end-user application rather than low-level building blocks
- your workload runs on CPU only without NVIDIA GPUs
- you need algorithms RAFT does not cover and would have to implement them yourself anyway

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, search-engine, math, gpu-computing, data-science
- domain: machine-learning, gpu-computing, data-science
- platform: cpp, python, cross-platform
- tags: cuda, nearest-neighbors, linear-algebra, sparse-operations, rapids, vector-search, primitives, header-only, search, algorithms, gpu, linux

## Member repositories
- NVIDIA/raft (main) score 94

## Provenance
- Observed fields: from GitHub, fetched 2026-09-01T02:13:55.828783+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T07:04:25.626820+00:00, confidence not recorded.
  - readme: https://github.com/NVIDIA/raft (fetched 2026-09-01T02:13:55.828783+00:00, sha 8be13f899f85)
  - homepage: https://docs.rapids.ai/api/raft/stable/ (fetched 2026-08-29T13:05:19.613112+00:00, sha da33ae21608f)
  - site_page: https://docs.rapids.ai/api/raft/stable/build (fetched 2026-08-29T13:05:19.622588+00:00, sha a23d2b89ddcb)
- Data as of 2026-08-30T08:39:29.467469+00:00.