Hugging Face

The default platform for sharing open-source AI models, datasets, and demos — the "GitHub of machine learning". Hosts millions of models including Llama, Qwen, DeepSeek, and Mistral.

Hugging Face is the New York and Paris-based company that runs the namesake platform — the default place to upload, download, and discover open-source AI models, datasets, and demos. Founded in 2016 originally as a chatbot app, the company pivoted into the open-source ML infrastructure space and became indispensable to the field. It matters because Hugging Face is where the open-source AI ecosystem actually lives. Almost every open-weight LLM (Llama, Qwen, DeepSeek, Mistral, Yi, GLM, Phi, Gemma) is published on Hugging Face Hub. The transformers Python library — also from Hugging Face — is the standard interface for loading and running these models. Spaces hosts demos. Datasets hosts training data. Tens of thousands of new model versions are uploaded every week. Key products and projects: Hub (model and dataset hosting, like GitHub for ML), transformers library (the de facto Python framework), datasets library, accelerate, PEFT (parameter-efficient fine-tuning), TRL (training reinforcement-learning post-training), Inference Endpoints (managed model hosting), Spaces (Gradio-based demos), and the Open LLM Leaderboard. The community-first culture and freemium model — free for individual use, paid for enterprise — has made it culturally central to AI. Without Hugging Face, the open-source LLM movement would look very different. Related: open-source, transformers library, ModelScope, GitHub.