fix: remove legacy conv converter

[chohk88]

Description

A RuntimeError occurs in many models when the output_padding argument for deconvolution is non-zero. The current CTX converter cannot handle this case, so the legacy FX converter is used instead. However, the legacy FX converter also raises a RuntimeError for deconvolutions with non-zero output_padding. If the legacy converter is removed, it results in graph breaks but avoids RuntimeErrors.

The ideal solution would be to implement a dedicated converter for deconvolutions with non-zero output_padding. However, TensorRT's Python API does not currently support output_padding as an input for tensorrt.IDeconvolutionLayer, making this implementation very challenging. It is recommended to create a separate issue to discuss and address this limitation.

Error message:

  File "/usr/local/lib/python3.12/dist-packages/model_navigator/commands/execution_context.py", line 156, in _execute_function
    fire.Fire(func, unwrapped_args)
  File "/usr/local/lib/python3.12/dist-packages/fire/core.py", line 143, in Fire
    component_trace = _Fire(component, args, parsed_flag_args, context, name)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/fire/core.py", line 477, in _Fire
    component, remaining_args = _CallAndUpdateTrace(
                                ^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/fire/core.py", line 693, in _CallAndUpdateTrace
    component = fn(*varargs, **kwargs)
                ^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/model_navigator/commands/convert/converters/ep2torchtrt.py", line 142, in convert
    with TimingCacheManager(model_name=model_name, cache_path=timing_cache_dir) as timing_cache:
  File "/usr/local/lib/python3.12/dist-packages/model_navigator/frameworks/tensorrt/timing_tactics.py", line 267, in __exit__
    raise exc_value
  File "/usr/local/lib/python3.12/dist-packages/model_navigator/commands/convert/converters/ep2torchtrt.py", line 149, in convert
    tr_model_compiled = torch_tensorrt.dynamo.compile(
                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch_tensorrt/dynamo/_compiler.py", line 288, in compile
    trt_gm = compile_module(
             ^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch_tensorrt/dynamo/_compiler.py", line 462, in compile_module
    trt_module = convert_module(
                 ^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch_tensorrt/dynamo/conversion/_conversion.py", line 142, in convert_module
    interpreter_result = interpret_module_to_result(
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch_tensorrt/dynamo/conversion/_conversion.py", line 121, in interpret_module_to_result
    interpreter_result = interpreter.run()
                         ^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch_tensorrt/dynamo/conversion/_TRTInterpreter.py", line 610, in run
    self._construct_trt_network_def()
  File "/usr/local/lib/python3.12/dist-packages/torch_tensorrt/dynamo/conversion/_TRTInterpreter.py", line 347, in _construct_trt_network_def
    super().run()
  File "/usr/local/lib/python3.12/dist-packages/torch/fx/interpreter.py", line 146, in run
    self.env[node] = self.run_node(node)
                     ^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch_tensorrt/dynamo/conversion/_TRTInterpreter.py", line 676, in run_node
    trt_node: torch.fx.Node = super().run_node(n)
                              ^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch/fx/interpreter.py", line 203, in run_node
    return getattr(self, n.op)(n.target, args, kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch_tensorrt/dynamo/conversion/_TRTInterpreter.py", line 783, in call_function
    return converter(self.ctx.net, target, args, kwargs, self._cur_node_name)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.12/dist-packages/torch_tensorrt/fx/converters/aten_ops_converters.py", line 125, in aten_ops_convolution
    raise RuntimeError(f"Target {target} does not support `transposed=True` ")
RuntimeError: Target aten.convolution.default does not support `transposed=True` 

While executing %convolution_15 : [num_users=1] = call_function[target=torch.ops.aten.convolution.default](args = (%cat, %model_1_submodule_1_submodule_1_submodule_2_0_conv_weight, %model_1_submodule_1_submodule_1_submodule_2_0_conv_bias, [2, 2], [1, 1], [1, 1], True, [1, 1], 1), kwargs = {_itensor_to_tensor_meta: {<tensorrt.tensorrt.ITensor object at 0x7f4269bc4770>: ((1, 1, 128, 128), torch.float32, False, (16384, 16384, 128, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269695370>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bc9b30>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bbb570>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bb8d30>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bbb630>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bb9bb0>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bb94b0>: ((16,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bb8130>: ((16,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a5a2b0>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a5a370>: None, <tensorrt.tensorrt.ITensor object at 0x7f4269a58130>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a58030>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a5a2f0>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269cdddf0>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269cdd7b0>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269cdf130>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269cdc0b0>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bba730>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a88130>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a5a4f0>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a9b0>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a3b0>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a2f0>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a370>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a170>: ((16,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a89fb0>: ((16,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a89d30>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a89df0>: None, <tensorrt.tensorrt.ITensor object at 0x7f4269a89bb0>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a881f0>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a89bf0>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a89930>: ((1, 16, 1, 1), torch.float32, False, (16, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a898f0>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a899f0>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a89830>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a894f0>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8aaf0>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269cdf930>: ((1, 16, 64, 64), torch.float32, False, (65536, 4096, 64, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a5a0b0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a591f0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a89430>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a88e30>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a893b0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269b52ef0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a88d70>: ((32,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a88a30>: ((32,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a887f0>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8ac70>: None, <tensorrt.tensorrt.ITensor object at 0x7f4269a89ff0>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b5b0>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bb9170>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b0b0>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a596f0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8bcf0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a58170>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a870>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b7f0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f426962b5f0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a8f0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b3b0>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b570>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a580b0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b1b0>: ((32,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a882f0>: ((32,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8ff0>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b90b0>: None, <tensorrt.tensorrt.ITensor object at 0x7f4269a89230>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bb8fb0>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b9370>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bc6b30>: ((1, 32, 1, 1), torch.float32, False, (32, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b9170>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b99f0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b92f0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b9c70>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a59d30>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a898b0>: ((1, 32, 32, 32), torch.float32, False, (32768, 1024, 32, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bba5b0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42688390b0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269694130>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8830>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b86f0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a883b0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8730>: ((64,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8570>: ((64,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8070>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a5a230>: None, <tensorrt.tensorrt.ITensor object at 0x7f42696bae70>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb0f0>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b170>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb030>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8ad30>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8270>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696babf0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a470>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a7b0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a6b0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b3f0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a88c70>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a88b30>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696ba570>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696ba430>: ((64,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b89b0>: ((64,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bbcb0>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bc4c30>: None, <tensorrt.tensorrt.ITensor object at 0x7f42696bb6f0>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb530>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b8f0>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb670>: ((1, 64, 1, 1), torch.float32, False, (64, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb5f0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a883f0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b370>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696baab0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb4b0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b85f0>: ((1, 64, 16, 16), torch.float32, False, (16384, 256, 16, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a896f0>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bbb6b0>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8ab30>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bbe70>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696ba2f0>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a0f0>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8ef0>: ((128,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb830>: ((128,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b81b0>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a830>: None, <tensorrt.tensorrt.ITensor object at 0x7f4269bbbeb0>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a430>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b85b0>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269cdcf30>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bbd30>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a89a30>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8aa70>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b9ef0>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a89af0>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8bab0>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a130>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8b8b0>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b82f0>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b9330>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8cb0>: ((128,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b9a70>: ((128,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a897b0>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb4f0>: None, <tensorrt.tensorrt.ITensor object at 0x7f42696ba4b0>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8c30>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269cdcfb0>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bbc30>: ((1, 128, 1, 1), torch.float32, False, (128, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bad70>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb130>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8230>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8a330>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bab30>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696149f0>: ((1, 128, 8, 8), torch.float32, False, (8192, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8bbf0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b9e30>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696145f0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269615ab0>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696157f0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb370>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bcb530>: ((256,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb2f0>: ((256,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b9670>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b9030>: None, <tensorrt.tensorrt.ITensor object at 0x7f4269a5a470>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb5b0>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696babb0>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696ba970>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269615bb0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a888f0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a89470>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696168b0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b90f0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a8abb0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb1b0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269a88c30>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269617b30>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696ba5f0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8cf0>: ((256,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269615030>: ((256,), torch.float32, False, (1,), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269617bb0>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b9430>: None, <tensorrt.tensorrt.ITensor object at 0x7f42696147f0>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb270>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bbb70>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696165f0>: ((1, 256, 1, 1), torch.float32, False, (256, 1, 1, 1), torch.channels_last, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696160b0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8430>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bcb6f0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696bb7b0>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269bc6770>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f4269615e30>: ((1, 256, 8, 8), torch.float32, False, (16384, 64, 8, 1), torch.contiguous_format, False, {}), <tensorrt.tensorrt.ITensor object at 0x7f42696b8a70>: ((1, 384, 8, 8), torch.float32, False, (24576, 64, 8, 1), torch.contiguous_format, False, {})}})
Original traceback:
  File "/usr/local/lib/python3.12/dist-packages/monai/networks/nets/unet.py", line 297, in forward
    x = self.model(x)
  File "/usr/local/lib/python3.12/dist-packages/monai/networks/layers/simplelayers.py", line 129, in forward
    y = self.submodule(x)
  File "/usr/local/lib/python3.12/dist-packages/monai/networks/layers/simplelayers.py", line 129, in forward
    y = self.submodule(x)
  File "/usr/local/lib/python3.12/dist-packages/monai/networks/layers/simplelayers.py", line 129, in forward
    y = self.submodule(x)

Fixes # (issue)

Type of change

Please delete options that are not relevant and/or add your own.

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

Checklist:

• My code follows the style guidelines of this project (You can use the linters)
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas and hacks
• I have made corresponding changes to the documentation
• I have added tests to verify my fix or my feature
• New and existing unit tests pass locally with my changes
• I have added the relevant labels to my PR in so that relevant reviewers are notified

[peri044]

If you plan to add this script, examples/dynamo would be a better fit. Please refer to the existing examples if you want to add this model as a part of model zoo.

[chohk88]

I added this script only to test deconv with output_padding. I’ve created a separate issue with the reproduction steps and code, so I’ll remove this code from this PR.

[zewenli98]

@chohk88 I think we currently already have conv and deconv converters https://github.com/pytorch/TensorRT/blob/main/py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py#L2467, but both didn't implement the functionality of output_padding as you mentioned. Is it possible to append constant_pad after the current conv/deconv implementation?

[chohk88]

@chohk88 I think we currently already have conv and deconv converters https://github.com/pytorch/TensorRT/blob/main/py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py#L2467, but both didn't implement the functionality of output_padding as you mentioned. Is it possible to append constant_pad after the current conv/deconv implementation?

@zewenli98 @peri044 I’ve added an example to this comment explaining how output_padding is supported using pre_padding and post_padding. In this PR, I removed the validator and added test cases for output_padding.

[zewenli98]

@chohk88 can you rebase onto the main branch?

[zewenli98]

LGTM!