Software Engineering Radio - the podcast for professional software developers Episode 549: William Falcon Optimizing Deep Learning Models
Feb 3, 2023
William Falcon, CEO of Lightning AI and creator of PyTorch Lightning, dives into optimizing deep learning models. He explains the distinctions between training and inference, emphasizing the need for fast processing in applications like self-driving cars. Falcon discusses MLOps complexities and highlights the required multidisciplinary skills for production AI. He also warns about common pitfalls for new users and the importance of using LightningModule for scalability. Plus, he shares insights on the evolving landscape of AI frameworks.
AI Snips
Chapters
Transcript
Episode notes
PyTorch + Lightning: Roles Defined
- PyTorch provides automatic differentiation and a flexible computational graph for building neural networks.
- Lightning organizes PyTorch code so teams can scale, adding structure and modularity for production workloads.
Lightning Used By Major AI Projects
- Major projects like Meta, NVIDIA's NeMo, and Stability AI's Stable Diffusion use Lightning in their training stacks.
- William Falcon cites public repos and companies to show broad production adoption of Lightning.
Optimization Extends Beyond Model Code
- Deep learning optimization spans data loading, algorithmic work, kernels and networking, especially at multi-node scale.
- Libraries like PyTorch and Lightning hide much complexity, but large-scale training still needs careful engineering.