The presentation discusses the journey of PyTorch, an open-source machine learning framework, and how the team focused on a specific market to excel and scaled their efforts while taking deliberate risks and measuring their progress.
- PyTorch focused on the ML researcher market and aimed to provide flexibility and debuggability in their framework
- The team took deliberate risks and made a bet on the future of the ML researcher market
- Measurement and metrics were important for PyTorch to track their progress and iterate on their product
- Scaling efforts required prioritizing and figuring out whether to vertically integrate or modularize the project
The team intentionally scoped down their efforts to focus on the animal researcher market and gradually expanded their scope as they grew and matured. They also had to figure out how to scale their efforts without falling back on measurement and metrics alone. One colleague read through all the information produced on the internet about PyTorch and prioritized feature development and bug fixes based on a subjective understanding of what was going on. However, this approach did not scale well after a certain point, and they had to figure out a better process to scale their efforts.