Company

Meet Hyperbolic CTO Yuchen Jin: Powering AI through Openness

XDiscordRedditYoutubeLinkedin

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

Dr. Yuchen Jin, Co-founder & CTO at Hyperbolic, is laser-focused on bringing open-access AI to the world. With a PhD in Computer Science in hand, and after building Apache TVM, one of the best AI compilers in the world, Yuchen is ready to give AI builders across the world access to the Compute and Inference tools he wished he had as a developer and researcher.

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.

We want to provide developers and researchers with the AI tools they need to help advance AI in a safe and open ways.

Dr. Yuchen Jin  •  Co-founder & CTO at Hyperbolic

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.

You never know what is really happening inside an AI model. But you know that it’s mimicking your biological neural network and it just works. It’s really amazing and there are tons of interesting problems you can study with it.

Dr. Yuchen Jin  •  Co-founder & CTO at Hyperbolic

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.

It was a eureka moment for me. It meant that if we could make it universally possible to run AI models on consumer grade hardware in way that was both performant and affordable, we could significantly increase access to supply for devs and researchers worldwide.

Dr. Yuchen Jin  •  Co-founder & CTO at Hyperbolic

Marrying AI with open-source development & blockchain

As someone who has open-sourced almost every research paper he’s every written, Yuchen is no stranger to sharing technology and breakthroughs that can help others. But his leadership role driving the Relax project at Apache TVM strikes a particular note with the open-source community and is a story that lives at the heart of web3.

While at OctoAI, Yuchen assembled a team of 6 engineers and led an open-source community that included top thinkers from AWS and leading chipmaker Qualcomm, as well as prominent researchers from CMU, UIUC, and UCSD. Together, they set about developing a next-generation graph-level language and core compilation infrastructure in Apache TVM, designed expressly to make it possible to run AI models on multiple chipsets. They called it Relax, although he has done quite the opposite ever since.

Relax, an innovation grounded in the spirit of collaboration and openness, laid a solid foundation for Hyperbolic’s commitment to open-source AI. It’s this trait that makes the company feel right at home in the open-source atmosphere that is the blockchain community, but it’s his alignment with Jasper that completes the picture.

“The partnership with Jasper is perfectly complementary. I have the background in AI and distributed systems, and he’s an expert mathematician who knows how to do proofs to verify if a node in the decentralized network is actually running the right model,” said Yuchen. “It’s a powerful combination that makes us perfectly suited to solving the core challenges inherent to decentralized 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 who have the talent and experience required to make AI consistently performant and affordable across a variety of chipsets. Even more importantly, he’s committed to make 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.

Blog
More Articles