Software from dynamic HPE and NVIDIA partnership makes generative AI a reality

[EIC_Alliances]

by Joey Zwicker - Co-Founder of Pachyderm

The race to adopt generative AI is here. Organizations of all sizes are rushing to develop large language models to create images, audio (music), and video. Unlike anything we have seen in the past, this rationale is grounded—AI can empirically move the needle—to reduce costs, increase productivity, and create a competitive advantage.

Currently, transformative technologies like generative AI are relegated to tech titans with deep pockets, infrastructure, and intellectual capital. These Fortune 10 organizations can leverage their position to invest in custom software, hardware, and developers to overcome the complexity and costs it takes to scale a PoC to production. This makes any impactful AI implementation wishful thinking for those without these resources.

Leveling the playing field

For years, NVIDIA has provided accelerated computing platforms and AI software that power machine learning. In the same vein, Hewlett Packard Enterprise servers (HPE ProLiant, HPE Apollo), HPE supercomputers (HPE Cray EX, HPE Cray XD), and HPE GreenLake have provided the data center infrastructure to operationalize AI. Now, HPE and NVIDIA are collaborating to offer customers a complete, end-to-end AI software stack to create custom, accurate, and trustworthy machine learning models. This solution is based on a validated technology stack, including NVIDIA AI Enterprise software, HPE Machine Learning Data Management Software, HPE Machine Learning Development Environment, and HPE Ezmeral for accelerated data centers.

At the core of the HPE AI software portfolio are HPE Machine Learning Data Management Software and the HPE Machine Learning Development Environment. These solutions are built on leading open source tools: Pachyderm and Determined AI, respectively. HPE Machine Learning Data Management Software is a flexible, scalable data pipelining and data versioning tool typically used for data preparation. The HPE Machine Learning Development Environment is a secure, cost-effective, distributed training solution with hyperparameter optimization and experiment tracking. These software solutions are offered with a reference architecture that encompasses all stages of the AI workflow from data preparation, model development, and training to deployment and monitoring.

NVIDIA AI Enterprise is the software layer of the NVIDIA AI platform, which offers over 100 frameworks, pretrained models, and development tools to accelerate data science and streamline development and deployment of production AI, including generative AI, computer vision, and speech AI. All aspects of the AI workflow are enhanced with NVIDIA RAPIDS for data processing, NVIDIA frameworks like NeMo for large language models and MONAI for medical data, as well as NVIDIA TensorRT and NVIDIA Triton for inference. As AI models and their development rapidly evolve and expand, the complexity of the AI stack and its dependencies grows. For a secure, optimized, full-stack solution designed to accelerate enterprises with support, security, and API stability, NVIDIA AI Enterprise offers enterprises a success path to the leading edge of AI without the potential risks of open source software.

NVIDIA and HPE software are complementary and work well together to drive organizational machine learning outcomes. This interchangeability (as shown in diagram below) empowers data science teams to combine the best HPE or NVIDIA software based on their expertise or use case. For example, HPE Machine Learning Data Management Software can leverage NVIDIA RAPIDS to accelerate data processing anywhere from 5x to 200x, depending on the workload, compared to CPU-only platforms.

One of the most popular workflows for generative AI today is fine-tuning large language models for specific use cases. The broad array of use cases ranges across summarization, sentiment analysis, QA, and search. The combined HPE and NVIDIA software stack, on top of HPE server infrastructure and NVIDIA accelerators, enables a powerful and seamless end-to-end experience. Starting with unstructured data processing, users can leverage HPE Machine Learning Data Management, which provides data versioning, lineage, and processing pipelines. Accelerated on GPUs with NVIDIA RAPIDS, these pipelines can run orders of magnitude faster compared to using CPUs. Once data is ready for training, the HPE Machine Learning Development Environment can manage NVIDIA accelerators, track experiments, and deliver advanced distributed training and hyperparameter searches. When these are paired with NVIDIA frameworks like NeMo and MONAI, users can produce state-of-the-art, fine-tuned models for horizontal use cases like conversational AI or vertical use cases specific to healthcare, manufacturing, and other key industries. At deployment, NVIDIA Triton can be used for optimized inferencing while tying in metadata from the source data set in HPE Machine Learning Data Management and the training configurations in the HPE Machine Learning Development Environment, resulting in full model lineage and establishing an end-to-end machine learning (ML) platform.

As part of deepening their collaboration, HPE is actively certifying its software to work with NVIDIA AI Enterprise. This rigorous testing demonstrates a higher level of performance, functionality, and confidence from knowing that the software is supported by both HPE and NVIDIA. The HPE Machine Learning Development Environment has passed certification. This initial certification will immediately be followed with testing of HPE Machine Learning Data Management Software. There is more to come, with additional certifications on the road map.

Summary

Generative AI is clearly a disruptive technology, spurring worldwide interest in large language models. Until now, the ability to access AI technology was limited to large organizations with deep resources. HPE and NVIDIA are making AI accessible to all with a complementary and end-to-end hardware, software, and services solution that enables organizations at any stage of AI maturity to create an AI advantage.

Take the next step and learn more about how HPE and NVIDIA can accelerate your AI initiatives from the edge to the cloud.

Joey Zwicker is the Co-Founder of Pachyderm where he leads all GTM and customer-facing functions of the business. Pachyderm was recently acquired by HPE and has been built into HPE’s AI at Scale portfolio of products and Joey leads the software sales, community, and portfolio strategy teams. In addition, Joey is one of the creators and on the Board of Directors of the AI Infrastructure Alliance (AIIA). AIIA is a non-profit organization focused on bringing together leaders in the AI, Machine Learning, and MLOps space to help develop standards and interoperability across tools and practitioners that drive AI advancements forward.