Tracel AI: A New Player in the Field of High-Performance Computing in Deep Learning

Presentation

Tracel is a tech company that develops deep learning infrastructures, optimized for both training and inference. We help researchers and engineers bring their vision to life by building the tools that enable them to create reliable and efficient models with reduced development time. We want our work to play a crucial role in democratizing the future of artificial intelligence and its applications across various industries, while contributing to the advancement of the field.

Products

As of this writing, our core endeavor is to build Burn: a new comprehensive dynamic Deep Learning Framework built using Rust with extreme flexibility, compute efficiency, and portability as its primary goals.

Burn is and always will be open-source, as this is the best way to increase its impact and achieve our mission. Tracel will be supporting businesses building on Burn by providing complementary tools and services that meet their needs.

Core Values

It is not surprising that one of our core values is transparency! We believe that sharing our journey with the community, the successes, but also the shortcomings, is important, especially in the field of artificial intelligence.

We also greatly value efficiency; in software, but also in everything we do. The world has limited resources and it's our job to make the most out of it! Relentlessly, we always strive to spend our time on the most important problems.

Finally, we are not perfect and errors are going to be made along the way. This isn't a problem by itself; we are only human after all. However, we strongly believe in continuous improvements and iteratively enhancing ourselves, as well as all of our work.

Background

Involved with open-source software since his bachelor's degree, Nathaniel always had a side passion project to experiment with his creative ideas. After working as a software engineer in the industry, Nathaniel becomes interested in Deep Learning and starts experimenting with it. He is so interested that he returned to school at MILA to pursue his search for knowledge.

After his Master's, he has to choose between academia or industry, and decides to join a startup in AI, where he thinks his software engineering and machine learning background could be the most useful. As always, Nathaniel has a passion project on the side. At that moment, he wants to experiment with asynchronous and sparse neural networks. However, he isn't able to materialize his idea because the available software tools are not flexible enough to render his vision.

Thinking he probably is not the only one, that many AI researchers must also feel constrained, he thinks the best way to concretize his idea would be to have access to better tooling. Meanwhile, at his job of the time, he experiences the pain of training and deploying models in production environments reliably and inexpensively.

Thus, he starts working on a deep learning framework written in Rust as his next side project. The choice of Rust as the programming language was made to better support safe concurrent programming and avoid the Python/C++ dichotomy that is ubiquitous in the field. He believes the general reason to use Rust is when one needs to go through multiple abstraction boundaries without having to pay for performance, which is the exact setting in which deep learning lies.

After some time working on the project, he decides to announce it online. The project receives hundreds of GitHub stars within days and attracts some serious contributors who help him improve the framework.

From then on, the idea of creating a startup around the project starts simmering in his head.

Founding the company

A few months later, Nathaniel easily convinces his longtime friend Louis to drop out of his Ph.D. in AI to embrace the exciting life of building a business. As Louis had already founded a tech startup in the past, his experience made it a no-brainer for him to join Nathaniel as a co-founder.

Discussing the commencement of the startup in a restaurant in their hometown, Quebec City, Nathaniel and Louis are brainstorming for a company name. From their table, the window offers a direct view of the “Tracel de Cap-Rouge,” a very rusty infrastructure supporting trains. The word Tracel, which is a Québécois deformation of the English word trestle, conveys the values of reliability and sustainability they wanted for their company, along with subtle references to both the language Rust and the hometown and mother tongue of the founders.

It was decided to name the company Tracel, and Nathaniel and Louis started working full-time on the project from this day onward.

Conclusion

Working at Tracel means working on the fundamentals of artificial intelligence, from scratching GPU algorithms to building state-of-the-art AI models. If you are passionate about deep learning, algorithm efficiency, or the Rust programming language, don't hesitate to get in touch with us!

We are really excited to start our journey of supporting the community in adding more intelligence to the world. The vision of Tracel goes beyond Burn, it is only a starting point and you can expect more from us in the future. Let's simplify the process of creating intelligence, together as a community.