Dell Technologies Inc. said today it’s partnering with Hugging Face Inc. to help enterprises deploy generative AI models on-premises and get their proof-of-concepts up and running in production more easily.

Until now, big cloud providers such as Microsoft Corp. and Google Cloud have largely dominated the generative AI industry, because they provide easy access to the powerful computing resources required to run and fine-tune large language models. However, enterprises, which are looking to customize LLMs and put them to better use in their own businesses, are wary of the costs associated with cloud. They’re also concerned about data security, so many view a hybrid approach to generative AI as much more viable.

In an interview with VentureBeat, Matt Baker, Dell’s senior vice president of AI strategy, added that customers have also reported frustration with the complexity associated with building AI applications. “If you recall with the era of big data, there was a real challenge progressing from proof-of-concept to production,” he said.

It’s the same story with generative AI, with which enterprises need a lot of help. Dell said its partnership with Hugging Face, which operates a platform for hosting open-source artificial intelligence projects, will enable more AI models to be deployed where the essential data they need lives. That’s mostly in on-premises systems, rather than the cloud. To enable this, Dell has been busy creating validated designs for on-premises systems that will support AI workloads, including storage, servers and accelerators.

The two companies say they have collaborated to create a portal where Dell customers can select from a library of open-source AI models that has been curated for performance, accuracy and use cases. They’ll be able to choose the most appropriate model for their use cases, select the optimized infrastructure to run it on-premises, and then get help in fine-tuning those models as well.

The Hugging Face Dell portal will feature custom, dedicated containers and scripts to help deploy them on Dell servers and data storage systems.

Dell Chief AI Officer Jeff Boudreau said the company is partnering with Hugging Face to offer customers the freedom to use open-source generative AI models with the peace of mind that comes from the reliability and security of on-premises systems. “This collaboration translates into enterprises being able to modernize faster by more simply deploying customized GenAI models powered by trusted Dell infrastructure,” he added.

In addition, the portal will offer access to various datasets, libraries and tutorials for generative AI model training. There will be templates and models designed to achieve very specific outcomes. Enterprises will be able to couple them with their own proprietary data within a software container and fine-tune it to their needs.

The fine-tuning of AI models for maximum performance is done using a technique known as retrieval augmented generation or RAG, which can be very time-consuming. RAG incorporates external knowledge sources to enhance generative AI models with supplementary information.

Using this method, users can create stepwise instructions for many different tasks. Dell said it will simplify the RAG process for customers with the use of a containerized tool based on the popular parameter-efficient techniques LoRA and QLoRA.

Through the Hugging Face Dell portal, customers will initially be able to select various Dell PowerEdge servers that have been designed for AI models. Later, the company will also add its workstations as deployment options. Dell’s Apex service, which is a portfolio of hardware products offered to customers on an as-a-service-basis, is also included.

Constellation Research Inc. analyst Holger Mueller said AI workloads are no longer running exclusively on the cloud, so it makes perfect sense for Dell to partner with the company that offers the largest selection of open-source LLMs around.

“The move isn’t a surprise, but what is surprising is why it took Dell so long to arrange this partnership,” Mueller said. “Dell has made big investments in AI infrastructure, but it remains to be seen how much hardware enterprises will buy for their on-premises AI workloads. Much may depend on how useful the preconfigured set of AI libraries for Dell’s platforms are.”

Hugging Face offers access to thousands of open-source LLMs, and has been focused on building partnerships with AI hardware suppliers. Earlier this year, it said it was teaming up with Advanced Micro Devices Inc. to increase the number of LLMs that can run on the company’s new AI accelerator, the MI300 chip.

Image: Benzoix/Freepik