Machine Learning & Deep Learning Packages
Machine-learning and deep-learning frameworks and model libraries for training, fine-tuning, and serving models (PyTorch, scikit-learn, Transformers).
Packages in Machine Learning & Deep Learning
Kind1
Sort:Used by Most Apps
PyTorch
A Python-first tensor library with GPU acceleration and a dynamic, define-by-run autograd engine for building and training deep neural networks.
scikit-learn
Simple and efficient tools for machine learning and data analysis in Python.
tiktoken
OpenAI's fast, Rust-backed BPE tokenizer for counting and encoding text into the exact tokens GPT models see.
Datasets
One-line loading and fast, Arrow-backed processing for thousands of ML datasets
sentence-transformers
A Python library for computing state-of-the-art sentence, text, and image embeddings, rerankers, and sparse and multi-vector representations.
ONNX Runtime
Cross-platform, high-performance ML inference and training accelerator for ONNX models
Tokenizers
Fast, production-grade tokenizer implementations for NLP and LLMs, written in Rust
accelerate
A thin PyTorch wrapper that runs the same raw training script on CPU, multi-GPU, or TPU with mixed precision, DeepSpeed, and FSDP support.
vision
PyTorch's official library for datasets, pretrained models, and image transformations in computer vision.
sentencepiece
A fast, language-independent subword tokenizer and detokenizer trained directly from raw sentences, purpose-built for neural network text generation and LLM pipelines.
opencv-python
Pre-built, pip-installable OpenCV bindings that give Python programs full computer vision and image-processing capabilities without compiling from source.
safetensors
A safe, zero-copy binary format for storing and loading ML tensors without pickle
PEFT
Hugging Face's library of state-of-the-art parameter-efficient fine-tuning methods for large models.
torchaudio
GPU-accelerated audio processing library built on PyTorch, with differentiable transforms, Kaldi-compatible features, and pretrained speech models.
spaCy
Industrial-strength Natural Language Processing library for Python
tesseract.js
Pure JavaScript OCR library that recognizes text in over 100 languages, running in the browser or Node.js via WebAssembly.
onnxruntime-node
Microsoft's official Node.js binding for running ONNX model inference in production JavaScript apps
Transformers.js
Run Hugging Face Transformers models directly in the browser or Node.js with no server required.
einops
A readable, framework-independent notation for tensor rearrangement, reduction, and repetition.
Whisper
A general-purpose speech recognition model for transcription, translation, and language identification, trained on large-scale weak supervision.
TRL
Fine-tune and align language models with SFT, DPO, GRPO, and RLHF trainers built on Hugging Face Transformers.
LanceDB
Serverless, embedded vector database for multimodal AI search
bitsandbytes
8-bit and 4-bit quantization primitives that make large language models fit and train on far less GPU memory.
ort
A safe Rust wrapper for ONNX Runtime — fast ML inference and training
