Easily Train Models with H100 GPUs on NVIDIA DGX Cloud

Hugging Face · 18 March 2024, 00:00

Hugging Face launched Train on DGX Cloud, a service allowing Enterprise Hub organizations to fine-tune AI models using NVIDIA H100 GPUs through a no-code interface integrated into the Hugging Face Hub. The service charges $8.25 per GPU hour for H100 instances, with an example showing that fine-tuning Mistral 7B on 1,500 samples costs approximately $0.45. Users can now access GPU compute on-demand without writing training scripts, with fine-tuned models automatically saved to private repositories. (Note: The service was deprecated as of April 10, 2025.)