01.py

```python from typing import List import torch

def my_compiler(gm: torch.fx.GraphModule, example_inputs: List[torch.Tensor]): print(">>> my_compiler() invoked:") print(">>> FX graph:") gm.graph.print_tabular() print(f">>> Code:\n{gm.code}") return gm

@torch.compile(backend=my_compiler) def foo(x, y): return (x + y) * x

if name == "main": a, b = torch.randn(10), torch.ones(10) foo(a, b)

another way to do this is to use torch logs

import torch

@torch.compile def foo(x, y): return (x + y) * x

x = torch.randn(10) y = torch.ones(10) foo(x, y)

TORCH_LOGS=graph_code python3 dynamo/01.py ```