NVIDIA and Hugging Face have announced the launch of Train on DGX Cloud, a new service that makes it easy to use open models with the accelerated compute infrastructure of NVIDIA DGX Cloud. This service is available to Enterprise Hub organizations and allows users to fine-tune popular Generative AI models like Llama, Mistral, and Stable Diffusion in just a few clicks within the Hugging Face Hub.

The new experience expands upon the strategic partnership between NVIDIA and Hugging Face announced last year to simplify the training and deployment of open Generative AI models on NVIDIA accelerated computing. One of the main problems developers and organizations face is the scarcity of GPU availability, and the time-consuming work of writing, testing, and debugging training scripts for AI models.

What Happened

Train on DGX Cloud offers an easy solution to these challenges, providing instant access to NVIDIA GPUs, starting with H100 on NVIDIA DGX Cloud. In addition, Train with DGX Cloud offers a simple no-code training job creation experience powered by Hugging Face AutoTrain and Hugging Face Spaces.

Enterprise Hub organizations can give their teams instant access to powerful NVIDIA GPUs, only incurring charges per minute of compute instances used for their training jobs. The service is currently available to support the following model architectures: Llama, Falcon, Mistral, Mixtral, T5, Gemma, Stable Diffusion, and Stable Diffusion XL.

Background and Context

NVIDIA DGX H100 is a high-performance computing infrastructure that allows data centers and research labs to train, deploy, and run artificial intelligence (AI) workloads and machine learning (ML) models more rapidly. The system features 8 NVIDIA H100 Tensor Core GPUs, providing significant acceleration for matrix operations, commonly used in neural networks.

The DGX H100 architecture is designed for high-performance AI and ML workloads, with each system including 8 NVIDIA H100 Tensor Core GPUs, NVLink switch systems, and high-speed networking. This allows multiple DGX H100 systems to work together as a cluster, accelerating workloads by distributing tasks across machines seamlessly.

Why it Matters

The launch of Train on DGX Cloud is significant for the adult industry because it addresses one of the main challenges faced by developers and organizations: the scarcity of GPU availability. With this new service, users can access powerful NVIDIA GPUs in just a few clicks, without having to worry about writing, testing, and debugging training scripts.

This will enable faster development and deployment of Generative AI models, which are increasingly being used in the adult industry for tasks such as content generation and moderation. The service also offers a simple no-code training job creation experience, making it easier for users to fine-tune popular Generative AI models like Llama, Mistral, and Stable Diffusion.

What Comes Next

The launch of Train on DGX Cloud is just the beginning. NVIDIA and Hugging Face are working together to optimize inference and make accelerated computing more accessible to the Hugging Face community. This includes leveraging their collaboration on NVIDIA TensorRT-LLM and optimum-nvidia, as well as making some of the most popular open models on Hugging Face available on NVIDIA NIM microservices.

For those interested in learning more about Train on DGX Cloud, a step-by-step tutorial is available on the Hugging Face Hub. The service is also being showcased at GTC this week, with a session on Wednesday 3/20 at 3pm PT where Jeff will guide users through Train on DGX Cloud and more.

Key Facts

  • Train on DGX Cloud is a new service that makes it easy to use open models with the accelerated compute infrastructure of NVIDIA DGX Cloud.
  • The service is available to Enterprise Hub organizations and allows users to fine-tune popular Generative AI models like Llama, Mistral, and Stable Diffusion in just a few clicks within the Hugging Face Hub.
  • Enterprise Hub organizations can give their teams instant access to powerful NVIDIA GPUs, only incurring charges per minute of compute instances used for their training jobs.
  • The service currently supports the following model architectures: Llama, Falcon, Mistral, Mixtral, T5, Gemma, Stable Diffusion, and Stable Diffusion XL.
  • Train on DGX Cloud offers a simple no-code training job creation experience powered by Hugging Face AutoTrain and Hugging Face Spaces.