library,category,description
Transformers,models,Pretrained ML models
Datasets,datasets,Access and preprocess datasets of any type and size for model training
Tokenizers,tokenization,Rust-based implementations of the most common tokenization algorithms for research and production
Diffusers,models,Pretrained diffusion models
pytorch-image-models (timm),models,Pretrained image models
LeRobot,models,Pretrained robotics models and datasets
Sentence Transformers,models,Pretrained text and image embedding models
Accelerate,training,Unified distributed training framework
PEFT,training,Parameter efficient fine-tuning methods for efficiently adapting large pretrained models to downstream tasks
TRL,training,Reinforcement learning methods for training transformer models
SetFit,training,Few-shot fine-tuning methods for Sentence Transformers
nanotron,training,Lightweight 3D parallelism pretraining for transformer models (production)
picotron,training,Lightweight 4D parallelism pretraining for transformer models (educational)
Optimum,optimization,Optimized training and inference for Transformers on specific hardware
Quanto,quantization,Quantize model weights and activations
bitsandbytes,quantization,8 and 4-bit model quantization
Text generation inference,serving,Deploy and serve high-performance large language models for inference
Text embeddings inference,serving,Deploy and serve high-performance text embedding models for inference
Gradio,ML apps,Build and share ML apps
Argilla,datasets,Collaboration tool for building high-quality datasets
Distilabel,datasets,Framework for generating synthetic data and AI feedback
candle,ML framework,Rust-based ML framework for enabling serverless inference
smolagents,agents,Build and run agents
transformers.js,ML framework,JavaScript implementation of Transformers for browsers
Safetensors,models,New format for safely storing tensors
huggingface_hub,Hub,Python client for the Hugging Face Hub
huggingface.js,Hub,JavaScript libraries for interacting with the Hub API
Lighteval,evaluation,Evaluate LLMs across multiple backends