A developer who wants to try an open-weights model usually starts on the Hugging Face Hub. Each model has a page with its files, license, a description of how it was trained and often a live demo. A few lines of code with the Transformers library download the weights and run the model on a laptop or a cloud GPU, usually through PyTorch. The same works for datasets, and “Spaces” host small web apps that show a model in action.
That makes the Hub the main distribution point for open-source models: when a lab releases open weights, they typically appear there. Companies also use private Hub accounts to share models internally.
Treat a download like any other third-party code. Anyone can upload, so check who published a model, its license, its documentation and whether it has been evaluated on tasks like yours. Model files can carry security risks, and popularity is not proof of quality.
Guests mention it as standard infrastructure. In episode 43 at 35:14, Liquid AI’s Maxime Labonne says a new model architecture has to be reimplemented “in every library that you want to use,” starting with Hugging Face Transformers. In episode 47 at 38:38, Google DeepMind’s Paige Bailey points listeners to open robotics models “available on Hugging Face.” Episode 47 was sponsored by Galileo (sponsor thanks at 49:12).
Hear it from the guest
“So you have to reimplement it from scratch in every library that you want to use. Hugging Face transformers, it can be VLLM, it can be SGLANG, LAMA CPP, etcetera.”
“For some of our open robotics models that are available on Hugging Face if folks wanna check them out.”
Quotes lightly edited to remove filler words.
Sources
- Hugging Face Hub documentation — Describes the Hub as a platform for open machine learning hosting over 2 million models, 1.5 million datasets and 1.5 million AI apps (Spaces), also used by private teams; company-reported counts; checked October 24, 2026.
- Wolf et al.: HuggingFace's Transformers: State-of-the-art Natural Language Processing (2020) — Paper introducing the open-source Transformers library, a unified API for pretrained transformer models; checked October 24, 2026.
Go deeper
From the conversation
-
Beyond Transformers: How Liquid AI Is Rethinking LLM Architecture | Maxime Labonne -
Gemini 3 & Robot Dogs: Inside Google DeepMind's AI Experiments | Paige Bailey -
Parenting Your AI Agents for Best Results | Cisco's Jeetu Patel -
Why Context Alone Isn't Enough for Enterprise AI Agents | WisdomAI CPO Kapil Chhabra -
Switching Models Shouldn't Mean Starting Over | Walrus Protocol's Kimberly Logan -
Tuning GPU Performance with AI Agents | AMD’s Anush Elangovan on ROCm 10 -
Every AI Agent Has an Evaluation Gap | Alex Ratner, Snorkel AI -
Why Enterprises Need a Different Approach to AI Agents | Lyzr’s Siva Surendira