#213213th of 242 in ML frameworks and libraries

implicit

Python library for recommender systems on implicit feedback data

Stars
3.8k
License
MIT
Last commit
May 2026
Last release
May 2026
Language
Python
Stars, last 29 weeks+72
25 Apr 2026: 3,772 stars2 May 2026: 3,774 stars9 May 2026: 3,774 stars16 May 2026: 3,774 stars23 May 2026: 3,780 stars30 May 2026: 3,786 stars6 Jun 2026: 3,789 stars13 Jun 2026: 3,790 stars20 Jun 2026: 3,792 stars27 Jun 2026: 3,794 stars4 Jul 2026: 3,796 stars11 Jul 2026: 3,798 stars18 Jul 2026: 3,799 stars25 Jul 2026: 3,799 stars1 Aug 2026: 3,801 stars8 Aug 2026: 3,807 stars15 Aug 2026: 3,810 stars22 Aug 2026: 3,810 stars29 Aug 2026: 3,811 stars5 Sept 2026: 3,814 stars12 Sept 2026: 3,817 stars19 Sept 2026: 3,822 stars26 Sept 2026: 3,822 stars3 Oct 2026: 3,825 stars10 Oct 2026: 3,828 stars17 Oct 2026: 3,830 stars24 Oct 2026: 3,833 stars31 Oct 2026: 3,835 stars7 Nov 2026: 3,840 stars14 Nov 2026: 3,844 stars

3,772 on 25 Apr 20263,844 on 14 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

Implicit provides Python implementations of ALS, Bayesian Personalized Ranking, Logistic Matrix Factorization and item-item nearest neighbour models (cosine, TFIDF, BM25) for implicit feedback datasets. Training is multi-threaded on CPU via Cython and OpenMP, and ALS and BPR also have custom CUDA kernels. Annoy, NMSLIB or Faiss can be plugged in to speed up recommendation lookups.

Who it is for: Python developers building recommenders from clicks, plays or purchases

Strengths

  • Multi-threaded CPU training using Cython and OpenMP across all cores
  • Custom CUDA kernels for ALS and BPR on GPU
  • Prebuilt pip wheels for x86_64 Linux, Windows and OSX; conda packages available
  • Optional Annoy, NMSLIB or Faiss integration for faster recommendations

Weaknesses

  • Limited to implicit-feedback collaborative filtering; no explicit-rating or deep-learning recommenders
  • GPU support requires CUDA Toolkit 13 and a separate rmm-cu13 install
  • Prebuilt wheels listed only for x86_64; other architectures not mentioned
  • CPU speed depends on tuning BLAS threads, such as OPENBLAS_NUM_THREADS=1

What it needs

  • GPU optional
  • Needs SciPy, NVIDIA CUDA Toolkit 13 (GPU only), rmm-cu13 (GPU only), Annoy/NMSLIB/Faiss (optional)
  • Models: Alternating Least Squares, Bayesian Personalized Ranking, Logistic Matrix Factorization, Item-item nearest neighbours (cosine, TFIDF, BM25)

What changed

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Also in ML frameworks and libraries

See all 242
Also in ML frameworks and libraries
RankProjectScore
1transformersModel-definition library for pretrained text, vision, audio and multimodal models169.3k starsApache-2.094 out of 100
2ultralyticsPython library and CLI for training and running YOLO vision models63.5k starsAGPL-3.094 out of 100
3pytorch-image-modelsPyTorch library of image models, pretrained weights and training scripts37.3k starsApache-2.094 out of 100
#4pytorchPython tensor library with GPU support and tape-based autograd105.8k starscustom license93 out of 100
#5supervisionPython toolkit for loading, annotating and counting computer vision detections Live demo ↗ (opens in a new tab)51.7k starsMIT92 out of 100