Meet Hyperbolic CTO Yuchen Jin: Powering AI through Openness

This post originally appeared on the Hyperbolic Medium blog on May 3rd, 2024

Hyperbolic believes that anybody with an idea and the will to make the world better should have open access to AI compute and inference; and Yuchen’s life experience is the very embodiment of this idea.

His path to Co-founder and CTO of Hyperbolic began with a early passion for computers and AI. He received his first computer from his father at the age of 10 and immediately began obsessing over online strategy games. He soon learned to code and that primed him for his a Bachelor’s in Computer Science and Engineering at the Huazhong University of Science & Technology in China.

Yuchen’s innate technical abilities helped him stand out and he immediately began placing at the top of his department. At age 20 he was handpicked to join a research group that included Ph.D. and Master’s students and went on to publish his first research paper at the age of 22. For his efforts, Yuchen was rewarded with the prestigious China National Scholarship as well as a MediaTek Scholarship.

Yuchen dives headfirst into AI during his Ph.D

Having seen firsthand what Ph.D. students get up to while still and undergraduate, Yuchen decided to pursue his CS Ph.D. at the University of Washington. This is when AI began to overtake the lectures, hallway conversations, and Yuchen’s imagination.

Under the supervision of Professor Arvind Krishnamurthy, ACM fellow and Vice President of USENIX — an association that researches distributed systems and computer networks — Yuchen began to explore and eventually became expert in machine learning systems, computer networks, and distributed systems. All topics that would prove pivotal to his future role at Hyperbolic.

Unlike most, Yuchen’s Ph.D. research was not merely an academic exercise. Rather, it laid the foundation for understanding complex AI architectures and networked systems that featured heavily in his commercial exploits. An excellent example of this was the development of the Automatic-learning Rate Scheduler during his tenure in ByteDance’s AI Lab, which significanlty accelerated their training of AI models. This project, and others like it, gave Yuchen invaluable hands-on exposure to the real challenges facing AI developers.

While completing his PhD, Yuchen actively looked for practical ways to apply his research and completed three commercial internships that gave him deeper exposure to AI and systems engineering. He also contributed to two separate independent research groups focused on building AI models and systems for Microsoft’s Azure.

A breakthrough moment for Hyperbolic

In the early days of Hyperbolic, Yuchen and his teammates were suffering from the same constraint every AI developer had to contend with at that time — limited and costly access to GPU Compute. This was nothing new to Yuchen, who had been dealing from the same problem as a PhD student, and later, in his role at Microsoft Research. Like most, he had come to accept that access to Nvidia GPU compute, was an unavoidable hurdle for AI developers.

But that all changed one day when Yuchen read a seminal blog post from the TVM community and OctoAI (Yuchen’s previous employer), the organization well-known by those in the field for the work they do to run, tune, and scale the models that power AI applications. This particular post detailed how the Llama2–7B/13B LLM could be efficiently run on a consumer-grade AMD GPU for less cost and with similar performance.

As someone who has open-sourced much of his research, Yuchen has long believed in building technology that others can use, study, and improve. His leadership on the Relax project within Apache TVM reflects that commitment to open development and technical collaboration.

While at OctoAI, Yuchen assembled a team of six engineers and helped lead an open-source community that included contributors from AWS, Qualcomm, CMU, UIUC, and UCSD. Together, they developed a next-generation graph-level language and core compilation infrastructure in Apache TVM, designed to make it easier to run AI models across multiple chipsets.

Relax, built through collaboration across the open-source AI community, laid a strong foundation for Hyperbolic’s commitment to making AI infrastructure more open, performant, and accessible.

“The partnership with Jasper is perfectly complementary. I have the background in AI and distributed systems, and he’s an expert mathematician with deep experience in verification,” said Yuchen. “It’s a powerful combination that makes us well suited to solving core infrastructure challenges in AI compute and inference.”

Hyperbolic’s future with Yuchen as CTO is exceedingly bright. He is one of only a handful of engineers in the world with the talent and experience required to make AI consistently performant and affordable across a variety of chipsets. Even more importantly, he’s committed to making AI more inclusive and accessible.

Fun facts about Yuchen:

• He worked alongside Ph.D’s and Master’s students when he was an undergrad, publishing his first research paper when he was 22.

• Led Relax, the next-generation compilation infrastructure in Apache TVM project, one of the best AI compilers in the world.

• Wrote the code for enterprise-grade infrastructure deployed in Microsoft Azure

To learn more about Dr. Yuchen Yuchen and his groundbreaking work with Hyperbolic, connect with him on X and join us on our quest to build a future where AI empowers humanity to its fullest potential.