About $1B of training compute is shared for free on the Hugging Face Hub (thomwolf-hub-dollars.static.hf.space)

Hugging Face has announced the release of approximately $1 billion worth of training compute for open-weight models on their Hub, marking a significant move in the AI/ML community. This estimate reflects the costs for GPUs used in final training runs for 75 official model families, showcasing how substantial computational resources were invested in the development of these models. By providing this compute for free, Hugging Face enables wider access for researchers and developers, fostering innovation and collaboration in AI technology. The significance of this initiative lies in its potential to democratize access to high-quality AI models, which were previously costly to train. The estimated costs are based on various GPU pricing models, with rates varying from $2 to $5 per GPU-hour, and the methodology provides a detailed breakdown, demonstrating a commitment to transparency. By sharing such resources, Hugging Face not only enhances research capabilities but also encourages the development of more refined and efficient AI systems, ultimately pushing the boundaries of what is possible in the field of machine learning.