Torch Backend in Transpiler

[yugborana]

PR Description

Related Issue

#28848

Checklist

• Did you add a function?
• Did you add the tests?
• Did you run your tests and are your tests passing?
• Did pre-commit not fail on any check?
• Did you follow the steps we provided?

[yugborana]

@Sam-Armstrong @YushaArif99 Good Afternoon. Would like to request for a review?

[Sam-Armstrong]

@yugborana Thanks for making the PR!

Looks like there's a few more things that will need changing before this works correctly. For example, you'll need to add torch as one of the SUPPORTED_S2S_TARGETS in ivy/transpiler/main_config.py.

I'd recommend using the following minimal example to check your solution is working correctly:

import kornia
import torch
import ivy

new_kornia = ivy.transpile(kornia, target="torch", reuse_existing=False)
rgb_image = torch.randn((1, 3, 224, 224))
gray_image = new_kornia.color.rgb_to_grayscale(rgb_image)

You can see this currently fails with the error ivy.transpiler.exceptions.exceptions.InvalidTargetException: 'target' must be one of ['tensorflow', 'jax', 'numpy', 'ivy', 'torch_frontend']. - but it looks like there are still other problems after fixing this.

[yugborana]

@Sam-Armstrong Thanks for your time and guidance. I have changed the targets list.

File "c:\Users\dell\ivy\test.py", line 7, in
gray_image = new_kornia.color.rgb_to_grayscale(rgb_image)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "c:\Users\dell\ivy\ivy\transpiler\main.py", line 375, in wrapper
callable_obj = transpile(
^^^^^^^^^^
File "c:\Users\dell\ivy\ivy\transpiler\main.py", line 772, in transpile
return translate(
^^^^^^^^^^
File "c:\Users\dell\ivy\ivy\transpiler\main.py", line 725, in translate
raise e
File "c:\Users\dell\ivy\ivy\transpiler\main.py", line 708, in translate
translated_object = translators_container.run_translators(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "c:\Users\dell\ivy\ivy\transpiler\translators_container.py", line 58, in run_translators
object = self.translators[-1].translate(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "c:\Users\dell\ivy\ivy\transpiler\translations\translator.py", line 402, in translate
object_like = self._setup_translation(object=object)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "c:\Users\dell\ivy\ivy\transpiler\translations\translator.py", line 115, in _setup_translation
self._output_dir = create_output_dir(
^^^^^^^^^^^^^^^^^^
File "c:\Users\dell\ivy\ivy\transpiler\utils\source_utils.py", line 247, in create_output_dir
_maybe_create_stateful_layers_module(NAME_GENERATOR.target, output_dir)
File "c:\Users\dell\ivy\ivy\transpiler\utils\source_utils.py", line 86, in _maybe_create_stateful_layers_module
stateful_file = os.path.join(output_dir, f"{module_name}.py")
^^^^^^^^^^^
UnboundLocalError: cannot access local variable 'module_name' where it is not associated with a value

Yeah many errors occurring.

[yugborana]

@Sam-Armstrong these are the errors, could you please guide more, where should I do changes?

[Sam-Armstrong]

@yugborana Can you push that change to the targets list? For this fix, it looks like you'll need to add another elif block in ivy/transpiler/utils/source_utils.py::_maybe_create_stateful_layers_module for torch where you set the module_name and source right?

[yugborana]

@Sam-Armstrong I have pushed the changes. But now, I think we have to do something with torch stateful layers?