compile

On my mac:

>>> torch.__version__
'2.2.2'
>>> torch.compiler.list_backends()
['cudagraphs', 'inductor', 'onnxrt', 'openxla', 'openxla_eval', 'tvm']

On Linux:

>>> torch.__version__
'2.6.0+cu118'
>>> torch.compiler.list_backends()
['cudagraphs', 'inductor', 'onnxrt', 'openxla', 'tvm']

TODO

• https://pytorch.org/docs/stable/torch.compiler_dynamic_shapes.html

• https://pytorch.org/docs/stable/torch.compiler_dynamo_deepdive.html