Hugging Face launches Train on DGX Cloud with NVIDIA H100 GPUs on HF Hub

AI Impact Summary

Train on DGX Cloud provides a no-code path to fine-tune open models on NVIDIA DGX Cloud via Hugging Face Hub, leveraging H100 GPUs and AutoTrain/Spaces for Enterprise Hub teams. The prior Train with DGX Cloud service is deprecated as of 2025-04-10, migrating users to a per-minute billing model with a Space-based workflow. Supported architectures include Llama, Falcon, Mistral, Mixtral, T5, Gemma, Stable Diffusion and Stable Diffusion XL; finished models are published to a private repository in the selected namespace.