Shriya Rishab grew up wanting to be a doctor or astronaut — fields where she could apply her perpetual curiosity and her passion for math and science to solve the world’s greatest challenges.
Today, as a senior deep learning engineer at NVIDIA, she’s doing just that.
After completing her master’s degree in computer science at Columbia University, New Jersey-based Rishab joined the company in 2020 with a focus on computer vision-based models and algorithms. After a few years, she became fascinated with the volumes of information communicated in text and shifted her expertise to large language models.
“Every time I learn something new, I become more and more curious. I’m always seeking new knowledge and trying to solve challenges in creative ways,” said Rishab, who also applies her creativity outside of work by painting and sketching.
For around two years, Rishab was part of the team leading NVIDIA’s submissions for MLPerf Training, a series of benchmarks designed to provide unbiased evaluations of how quickly systems can train models to a target quality metric. Twice a year, Rishab trained and submitted deep learning models to demonstrate the performance and capabilities of the NVIDIA AI platform.
The large language models she works with are the foundation for some of the most widely used AI services today. Rishab is especially proud of NVIDIA’s first time submitting a GPT-3-based model powered by NVIDIA H100 Tensor Core GPUs to MLPerf Training. The team broke MLPerf records, the result of many months of hard work from NVIDIA and other collaborators across the industry.
Rishab’s work created an opportunity for her to join MLCommons, the consortium of AI leaders that runs MLPerf. As the training cochair, she partners with representatives from academia, research and industry to help determine new, relevant models to add for future benchmarking.
“When we collaborate as a community, we can make things faster and better,” she said. “It’s important to me that MLCommons is an open, public benchmark. Anyone can view or submit, and it helps the whole industry learn and determine our future together.”
Rishab currently works on open-source, publicly available large language models and associated open-source platforms, such as NVIDIA NeMo, that anybody can run on NVIDIA hardware. She’s passionate about accelerating the most important models in the industry, while also making sure the training processes are resilient and efficient.
“Most of the work I’ve done at NVIDIA has been open source, and I’m proud of making AI more open and transparent,” she said. “We think about the whole community in everything we do. It’s proof of my life’s work.”
Learn more about NVIDIA life, culture and careers.