Today we announce the Dell Enterprise Hub, a brand new experience on Hugging Face to simply train and deploy open models on-premise using Dell platforms.
Try it out at dell.huggingface.co
Enterprises need to construct AI with open models
When constructing AI systems, open models is the perfect solution to fulfill security, compliance and privacy requirements of enterprises:
- Constructing upon open models allows corporations to know, own and control their AI features,
- Open models might be hosted inside enterprises secure IT environment,
- Training and deploying open models on-premises protects customers data.
But working with large language models (LLMs) inside on-premises infrastructure often requires weeks of trial and error, coping with containers, parallelism, quantization and out of memory errors.
With the Dell Enterprise Hub, we make it easy to coach and deploy LLMs on premise using Dell platforms, reducing weeks of engineering work into minutes.
Dell Enterprise Hub: On-Premise LLMs made easy
The Dell Enterprise Hub offers a curated list of essentially the most advanced open models available today, including Llama 3 from Meta, Mixtral from Mistral AI, Gemma from Google and more.
To access Dell Enterprise Hub, all you wish is a Hugging Face account.
The Dell Enterprise Hub is designed from the bottom up for enterprises, and optimized for Dell Platforms.
You may easily filter available models by their license or model size.
When you’ve chosen a model, you possibly can review a comprehensive model card designed for enterprise use. At a look you see key information in regards to the model, its size, and which Dell platforms support it well.
Many models from Meta, Mistral and Google require authorization to get access to the model weights. Because Dell Enterprise Hub is built upon Hugging Face user accounts, your account entitlements transfer over to Dell Enterprise Hub, and also you only must get permission once.
Deploy open models with Dell Enterprise Hub
When you’ve chosen a deployable model, deploying it in your Dell environment is admittedly easy. Just select a supported Dell platform, and the variety of GPUs you should use to your deployment.
While you paste the provided script in your Dell environment terminal or server, all the things happens automagically to make your model available as an API endpoint hosted in your Dell platform. Hugging Face optimized deployment configurations for every Dell platform, considering the available hardware, memory and connectivity capabilities, and repeatedly tests them on Dell infrastructure to supply the perfect results out of the box.
Train open models with Dell Enterprise Hub
Fantastic-tuning models improves their performance on specific domains and use cases by updating the model weights based on company-specific training data. Fantastic-tuned open models have been shown to outperform the perfect available closed models like GPT-4, providing more efficient and performant models to power specific AI features. Since the company-specific training data often includes confidential information, mental property and customer data, it is necessary for enterprise compliance to do the fine-tuning on-premises, so the info never leaves the corporate secure IT environment.
Fantastic-tuning open models on premises with Dell Enterprise Hub is just as easy as deploying a model. The fundamental additional parameters are to offer the optimized training container with the Dell environment local path where the training dataset is hosted, and where to upload the fine-tuned model when done. Training datasets might be provided as CSV or JSONL formatted files, following this specification.
Bring your Own Model with Dell Enterprise Hub
What if you should deploy on-premises your personal model without it ever leaving your secure environment?
With the Dell Enterprise Hub, when you’ve trained a model it’s going to be hosted in your local secure environment at the trail you chose. Deploying it’s just one other easy step by choosing the tab “Deploy Fantastic-Tuned”.
And if you happen to trained your model on your personal using considered one of the model architectures supported by Dell Enterprise Hub, you possibly can deploy it the very same way.
Just set the local path to where you stored the model weights within the environment you’ll run the provided code snippet.
Once deployed, the model is offered as an API endpoint that is straightforward to call by sending requests following the OpenAI-compatible Messages API. This makes it super easy to transition a prototype built with OpenAI to a secure on-premises deployment arrange with Dell Enterprise Hub.
We’re just getting began
Today we’re very excited to release the Dell Enterprise Hub, with many models available as ready-to-use containers optimized for a lot of platforms, 6 months after announcing our collaboration with Dell Technologies.
Dell offers many platforms built upon AI hardware accelerators from NVIDIA, AMD, and Intel Gaudi. Hugging Face engineering collaborations with NVIDIA (optimum-nvidia), AMD (optimum-amd) and Intel (optimum-intel and optimum-habana) will allow us to supply ever more optimized containers for deployment and training of open models on all Dell platform configurations. We’re excited to bring support to more state-of-the-art open models, and enable them on more Dell platforms – we’re just getting began!
