Why PyTorch Co‑Founder Soumith Chintala Is Leaving Meta After 11 Years

Announcement

Soumith Chintala, a co‑founder of PyTorch, announced that he will leave Meta on November 17. He said the decision was simple: he does not want to spend his entire career bound to PyTorch and feels it is the right time to start a new chapter.

Reasons for Leaving

He returned from a long vacation to find the project thriving without his direct involvement, which he sees as an ideal moment for transition. He also wants to explore opportunities outside Meta rather than staying in a comfortable senior role.

Impact of PyTorch

During his eleven‑year tenure at Meta, Soumith led PyTorch’s growth from a research prototype to a framework used by over 90% of AI practitioners worldwide, handling training tasks at the hundred‑billion‑parameter scale. The library powers foundational models across academia and industry, from MIT labs to startups.

Prominent figures such as Andrej Karpathy, Yann LeCun, and many AI leaders have praised PyTorch and Soumith’s contributions.

Farewell Letter

In his farewell note, Soumith expressed deep appreciation for the community, describing his time at Meta as the most rewarding part of his career. He highlighted the collaborative spirit, the joy of building tools, and the importance of open‑source in advancing AI.

He thanked the core team—Greg, Alban, Ed, Jason, Joe, John, Jana, and many others—for their resilience and shared values, and emphasized that the project will continue to thrive without his direct oversight.

Reflections on Meta

Soumith recalled his early days at FAIR, working with pioneers like Emily Denton, Rob Fergus, and Yann LeCun on GANs and other breakthroughs. He described the period as the most productive and enjoyable years of his professional life.

He also mentioned his involvement in robotics, AI infrastructure, and contributions to the Torch community, answering thousands of forum questions.

Personal Background

Born in Hyderabad, India, Soumith studied at NYU and VIT Vellore, joined Meta in 2014, and rose to a VP‑level role. His research spans object detection, generative modeling, game AI, and machine‑learning systems. He co‑authored influential papers such as LAPGAN, DCGAN, and Wasserstein GAN.

Future Plans

He plans to pursue smaller, novel projects that challenge him, emphasizing curiosity over staying within Meta. He remains committed to following PyTorch’s progress and may contribute occasional issues.

Gratitude

Soumith thanked countless colleagues, mentors, and community members who helped shape PyTorch, acknowledging their collective effort in turning a library into a global movement.

References

For more details, see his blog post and social media announcements.