Installer ngt avec Homebrew, Nix

Liens du projet

• Page d'accueil
• Docs amont
• Page du gestionnaire de paquets

Commandes et alias

• ngt
• qbg

Historique du projet

Masajiro Iwasaki wrote in April 2023 that Yahoo! JAPAN Research released NGT as open-source software in September 2016. The GitHub README credits Yahoo Japan Corporation from 2015 and anchors the project in graph/tree approximate nearest-neighbor methods, later adding ONNG, NGTQ, QG, and QBG variants.

The release notes embedded in the README show a progression from ONNG availability in August 2018, quantized NGT in December 2018, PyPI packaging in January and June 2019, QG in January 2021, FP16 in February 2022, and QBG/QG command consolidation in August 2022. Homebrew packaged stable version 2.7.4 on July 1, 2026.

Historique d'adoption

NGT's adoption is specialized but real in the vector-search world. The 2019 Opensource.com article says Yahoo! Japan used it for similarity-based fashion-item search, and the 2023 maintainer article says it was adopted by a number of Yahoo! JAPAN services through Vald. It is also cited in ANN benchmark and similarity-search literature.

Homebrew analytics on July 1, 2026 reported 268 install-on-request events over 365 days, which fits the package's role: useful for ML/search engineers and benchmarking people, not a mainstream developer utility.

Modes d'utilisation

The package-nerd path is to install `ngt`, generate or export vectors, create an index with a command such as `ngt create -d 300 -D c index objects.ssv`, and search with `ngt search -n 10 index query.ssv`. Python users can install the binding from PyPI under the `ngt` package and use `ngtpy`.

QG and QBG matter when datasets get large enough that memory traffic dominates. The README describes NGT as graph/tree based, QG as quantized graph based, and QBG as quantized blob graph based; QBG is specifically documented for billion-object scale.

Chronologie

• {'date': '2016-09', 'event': "Yahoo! JAPAN Research released NGT as open-source software, according to Masajiro Iwasaki's 2023 article."}
• {'date': '2018-08-08', 'event': 'ONNG became available in NGT v1.4.0.'}
• {'date': '2019-01-17', 'event': 'Python NGT became installable through PyPI in v1.5.1.'}
• {'date': '2021-01-15', 'event': 'NGT v1.13.0 added Quantized Graph support.'}
• {'date': '2022-08-10', 'event': 'QBG and renewed QG became available in v2.0.0.'}

Related projects

• Vald
• ANN-Benchmarks
• Faiss
• hnswlib
• Annoy
• ScaNN

Sources

• https://formulae.brew.sh/api/formula/ngt.json
• https://github.com/NGT-labs/NGT
• https://medium.com/@masajiro.iwasaki/fusion-of-graph-based-indexing-and-product-quantization-for-ann-search-7d1f0336d0d0
• https://opensource.com/article/19/10/ngt-open-source-library
• https://research.lycorp.co.jp/en/softwaredata