How Does PyTorch Compile Regular Python Code?
Topics covered in this part:
- Why can't a compiler optimize Python directly?
- What is a Computation Graph?
- Why do compilers need a Computation Graph?
- What is TorchDynamo?
- How does TorchDynamo capture Python execution?
- How is an FX Graph built?
- What does TorchInductor do next?
- How does the complete torch.compile pipeline work?