Convert exported model from TRAINING dialect to Aten dialect

1

I am trying to understand why exported_program from torch.export.export is having TRAINING dialect.

model.eval()
 with torch.no_grad():
     exported_model = torch.export.export(model, args)
     print(exported_model.dialect)

Model has no TRAINING related operations.
These are all the operations model has:

  1. placeholder
  2. call_function

Target for all the call_function is all Aten operations.How can I convert this exported program to Aten dialect ?

3

Hi @Himalayjor. Does exported_model.run_decompositions() work for you?

1 Like