Beyond the Hype: Domino Offers Production-Ready Generative AI Powered by NVIDIA

With the ongoing generative AI hype, one concept is becoming increasingly clear: giant, generic generative AI models, by themselves, are not the key to unlocking business value. While they are excellent for experimentation, entertainment, and some limited end-user work augmentation (ChatGPT might have helped with parts of this blog), they often fall short in terms of performance, accuracy, and risk when they aren't production grade.

Domino customers now have more options to leverage pre-trained foundation models from NVIDIA, which increases potential for production-ready generative AI at enterprise scale.

The release of NVIDIA AI Workbench — announced today at SIGGRAPH 2023 along with the NVIDIA AI Enterprise 4.0 software platform, which includes the NVIDIA NeMo framework — provides enterprises with production-ready generative AI through powerful new frameworks, management features, and tools for the development, fine-tuning, and deployment of large language models (LLMs).

Why is this especially exciting for Domino customers? Our collaboration with NVIDIA puts generative AI into practice by accelerating time-to-value while controlling the costs of building and operating responsible AI.

Transforming Beyond Generic AI: Building and Fine-Tuning Is the Future

“Data scientists believe generative AI will significantly impact enterprises over the next few years, but its capabilities cannot be outsourced — that is, enterprises need to fine-tune or control their own gen AI models,” summarized a recent VentureBeat article.

The article refers to Domino’s recent REVelate survey, finding that only 6% of AI professionals view commercial AI features (from ISVs and other third parties) as a viable strategy for a competitive advantage. The other 94% believe their organizations must create their own generative AI offerings. And most AI professionals (55%) plan to create differentiated customer experiences by fine-tuning foundation models from third parties rather than building their own - which requires more resources and technical know-how.

As part of Domino’s Summer ‘23 release, our AI Hub includes pre-built solutions and templates allowing enterprises to fine-tune and integrate these foundation models - automating transparent machine learning to accelerate AI projects customized to internal data and other specialized needs to control their own models.

Domino's Work with NVIDIA: Expanding on a Record of Enterprise AI Success

Whether by helping Lockheed Martin reduce infrastructure costs with a hybrid approach, scaling data science in a hybrid and multi-cloud computing environment at Johnson & Johnson, or putting models at the heart of Allstate’s business, Domino and NVIDIA’s collaboration has helped some of the world’s most advanced companies unleash AI at scale.

Since Domino’s NVIDIA AI Enterprise validation and integration in 2022, we have been continually building to help customers seamlessly deploy models in any environment - from data centers to the cloud to the edge - to simplify and accelerate the AI lifecycle. To meet the growing demand for data center-to-cloud flexibility, capacity, and cost-efficiency, Domino announced NVIDIA as the first launch partner for Domino Nexus - a single pane of glass that lets data science teams run AI workloads in any cloud, region, or on premises.

Domino natively supports the NVIDIA NGC catalog and NVIDIA AI platform, with two-way code interoperability between Domino, raw NCG containers, and NVIDIA AI Workbench - supporting rapid deployment, management, and scale of AI workloads across a variety of deployments.

The NeMo framework is a welcome addition to Domino's work with NVIDIA technologies. With NVIDIA AI Enterprise, which includes NVIDIA Nemo, data scientists can fine-tune LLMs in Domino’s platform for domain-specific use cases based on proprietary data and IP - without starting from scratch. This lowers the data, compute, training, and inference requirements for fine-tuned LLMs in domain-specific use cases across the hybrid- and multi-cloud.

Production Grade: Responsible, Cost-Effective Generative AI

While quick time-to-value for AI can deliver an incredible competitive advantage - a flawed model can wreak havoc on business. Domino’s REVelate survey revealed significant hurdles perceived by AI professionals, with challenges around governance (57%), bias and fairness (51%), and issues of control (49%).

NVIDIA AI Workbench creates a simplified path for developers to create AI-based applications from pre-trained models in just a few clicks, and Domino’s AI platform adds enterprise-ready governance functionality for the safe production and operationalization of models.

First, Domino Nexus unifies data science silos across the enterprise - across complex multi-cloud and on-premises architectures. Validated for use with the NVIDIA AI platform, Domino Nexus allows modern enterprises to run workloads where they make the most sense without infrastructure constraints. Enterprises can govern data access to protect data sovereignty, right-size cloud or on-premises infrastructure for AI workloads to optimize performance and costs, and future-proof infrastructure to support multi-cloud strategies and migrations.

Second, Domino’s recent innovations focus on two key responsible AI considerations: model risk/compliance and cost. Domino Model Sentry includes a unified model registry, complete model lineage tracking and audibility, and model approval and validation workflows. Cost governance capabilities help companies manage and reduce spend using fine-grained visibility, budgets, and quotas.

Domino, integrated with NVIDIA AI Enterprise for LLM development, provides an end-to-end enterprise MLOps platform - a unified system of record across the hybrid- and multi-cloud to help enterprises build and operate AI quickly, responsibly, and economically - without sacrificing collaboration or governance.

Domino in Practice with NVIDIA NeMo

Data scientists can leverage the NVIDIA NeMo, part of NVIDIA AI Enterprise, within Domino - adding the pre-built NeMo NGC catalog image to Domino’s self-serve development environment with on-demand access to data and powerful compute resources. Domino’s automatic compatibility adapts workspace tooling integrations with the NeMo toolkit so data scientists can access freely available, state-of-the-art pre-trained NeMo models on HuggingFace Hub and NVIDIA NGC. With just a few lines of code, these models can transcribe audio, synthesize speech, or translate text.

More advanced data science teams looking to fine-tune NeMo models can leverage sample scripts for multi-GPU/multi-node training. Domino automatically manages code and data versioning, while distributed auto-scaling clusters (including Spark, Ray, and Dask) dynamically grow and shrink based on workload demands.

Because Domino maintains two-way code interoperability with NVIDIA AI Workbench and NGC containers, teams can freely collaborate across platforms with ultimate flexibility. Models built in NVIDIA AI workbench can be deployed and monitored through Domino’s platform for a unified single pane of glass across hybrid- and multi-cloud environments. Conversely, models developed in Domino can be customized in NVIDIA AI Workbench.

Domino's collaboration with NVIDIA allows data scientists to focus on AI innovation and delivering business value rather than DevOps work - while data science leaders, IT platform owners, and business leaders can have confidence in efficient resource allocation/spend, governance, and risk mitigation.

Learn More: Cut Costs, Not AI Innovation

Join experts from Capgemini, Domino, and NVIDIA on August 29th at 9am PT, for a webinar discussing novel approaches that increase data science team productivity and impact while addressing the need to tighten belts. Whether you’re a data scientist, IT professional, or decision-maker, this webinar is your gateway to transformative strategies for smarter AI investments.

Register Now: Smarter Savings on Generative AI: Cut Costs, Not Innovation