The goal of the NVIDIA AI Technology Center (NVAITC) is to enable and accelerate AI research, education and adoption, using supercomputing resources based on NVIDIA technology, as well as to foster academic collaborations across the NVAITC network. A central objective of NVAITC collaborations is to provide support to specific research projects in AI and Applied AI, governed by an approved statement of work. The goal of projects is to foster technological development based on research, and to publish and otherwise disseminate results of the project’s work. NVAITC enables researchers to benefit from NVIDIA’s expertise in utilizing GPU and A I Computing. NVAITC projects enable academics at all levels to do their research more efficiently.

As researchers invest a lot of effort to unlock the full potential of AI, the need for scalable and efficient tools has become funamental, particularly when dealing with large models, diverse data formats, and high-performance systems. As NVAITC, we are working to facilitate this process by developing comprehensive tools, materials, and recipes that simplify AI adoption at scale.

The NVAITC Playbooks provide NVIDIA-based scalable reference implementations for common AI use-cases in research.

• multi-scale-agentic-rag-playbook: A playbook showcasing how to create a RAG pipeline working at different scales.
• synthetic-data-generation-and-sft-playbook: A playbook showcasing a scalable pipeline to finetune an LLM on synthetically enriched data through using the NeMo Framework.
• accelerated-video-for-ai-playbook: A playbook delving into HW and SW aspects that accelerate video processing for machine learning applications and proposes solutions on how to speed up dataloading with DALI and pyNvVideoCodec.
• hpc-ai-online-training-playbook: A playbook demonstrating how the perform online deep learning training and inference in a numerical simulation application using our TorchFort library.