The best machine learning libraries, according to hackers
19 machine learning libraries recommended in Hacker News comments over the past year, ranked by how many separate people recommend each one.
All machine learning libraries
- PyTorchThe de facto deep learning framework, praised for its Pythonic, debuggable design, easy GPU acceleration and huge ecosystem; critics cite installation headaches with GPU/CPU wheels and Python overhead.131recommend
- MLXApple's free machine learning framework, recommended for running models fastest on Apple Silicon Macs. Critics point to bugs and to its lack of Neural Engine or matrix hardware use.75recommend
- JAXGives NumPy-style array code automatic differentiation and JIT compilation, with fast GPU execution through XLA. Common caveat: the debugging story is rough compared to PyTorch.49recommend
- OpenCVFree, long-lived computer vision library that people find easy to start with and fast for image loading. Critics note uneven code and doubt the new engine.46recommend
- tinygradPraised for a small, readable deep learning framework that competes with PyTorch in far less code and reaches many hardware backends. Common caveat: little production use yet and slow inference.40recommend
- TensorFlowEarly end-user machine learning library, pivotal to fast AI development and widely used. Critics cite hard debugging, breaking changes between versions, and losing ground to PyTorch.20recommend
- ggmlSmall, portable C/C++ tensor library that people rely on to run local AI models, praised for its openness and quantization work.19recommend
- XGBoostRecommended as a strong, fast baseline for tabular data that often beats neural networks. Main caveat: it needs a lot of feature engineering.13recommend
- Hugging Face CandleRecommended as a Rust machine learning library for running models locally, with fast inference and easy model downloads. Caveat: critics call it immature and not yet feature complete.11recommend
- Burn deep learning frameworkRust deep learning framework that runs without Python, with autodiff and a libtorch backend option. Critics say it is still immature and not feature complete.11recommend
- jax-jsFree JavaScript machine learning library with much of Jax's API, running in the browser on WebGPU or WebAssembly. Critics find the API low-level and error-prone.9recommend
- Hugging Face TransformersCore open source Python library for running machine learning models, valued for smooth setup. Critics note frequent breaking changes and trouble working offline.8recommend
- Transformers.jsDevelopers recommend it as the go-to way to run machine learning models in the browser, with WebGPU support and many models since v4.8recommend
- scikit-learnPeople recommend it for evaluation, model selection and calibration tools, and for reliable training without manual tuning. Some would not use it for production code.7recommend
- LightGBMRecommended as a default for tabular data, often beating neural networks, though most people note it still needs good feature engineering.7recommend
- MediaPipeFree Google library for real-time hand and pose tracking from a plain webcam, often run in the browser for demos.6recommend
- Core MLPeople recommend it for running many models on Apple hardware, including the Neural Engine, but critics say it gives little control over which chip runs each layer.5recommend
- PyTorch LightningRecommended as a modern Python library with clear documentation for machine learning training. Some critics note a claimed copy of FastAI and a reported malicious 2.6.2 package.5recommend
- spaCyFree open-source NLP library, often much faster than API-based options, and it runs locally without third-party calls. Some users note it needs configuring.5recommend