MetaTensor conflict with Tensorboard

[ChenglongWang]

Describe the bug
TL,DR: Error occurred when using SummaryWriter.add_graph().

To Reproduce
Steps to reproduce the behavior:

1. The bug can be reproduced on 'segmentation2d' tutorial.
2. Just simply add writer.add_graph(model, inputs, verbose=True) in the training loop
3. SummaryWriter from either torch.utils.tensorboard or tensorboardX failed.

Environment (please complete the following information):

• OS: Ubuntu 18.04
• Python version: 3.10
• MONAI version: 1.1.0
• CUDA/cuDNN version: 11.6
• More details:

MONAI version: 1.1.0
Numpy version: 1.23.5
Pytorch version: 1.12.1
MONAI flags: HAS_EXT = False, USE_COMPILED = False, USE_META_DICT = False
MONAI rev id: a2ec3752f54bfc3b40e7952234fbeb5452ed63e3

Optional dependencies:
Pytorch Ignite version: 0.4.10
Nibabel version: 5.1.0
scikit-image version: 0.20.0
Pillow version: 9.4.0
Tensorboard version: 2.12.1
gdown version: 4.7.1
TorchVision version: 0.13.1
tqdm version: 4.65.0
lmdb version: 1.4.1
psutil version: 5.9.4
pandas version: 2.0.0
einops version: 0.6.0
transformers version: 4.21.3
mlflow version: 2.2.2

Exception messages:

generating synthetic data to /tmp/tmp9810tcm9 (this may take a while)
torch.Size([10, 1, 96, 96]) torch.Size([10, 1, 96, 96])
----------
epoch 1/10
1/5, train_loss: 0.3956
Traceback (most recent call last):
  File "/homes/clwang/Code/tests/add_graph_test.py", line 174, in <module>
    main(tempdir)
  File "/homes/clwang/Code/tests/add_graph_test.py", line 127, in main
    writer.add_graph(model, inputs, verbose=True)
  File "/homes/clwang/.miniconda3/envs/monai/lib/python3.10/site-packages/tensorboardX/writer.py", line 937, in add_graph
    self._get_file_writer().add_graph(graph(model, input_to_model, verbose))
  File "/homes/clwang/.miniconda3/envs/monai/lib/python3.10/site-packages/torch/utils/tensorboard/_pytorch_graph.py", line 338, in graph
    trace = torch.jit.trace(model, args, strict=use_strict_trace)
  File "/homes/clwang/.miniconda3/envs/monai/lib/python3.10/site-packages/torch/jit/_trace.py", line 750, in trace
    return trace_module(
  File "/homes/clwang/.miniconda3/envs/monai/lib/python3.10/site-packages/torch/jit/_trace.py", line 992, in trace_module
    _check_trace(
  File "/homes/clwang/.miniconda3/envs/monai/lib/python3.10/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "/homes/clwang/.miniconda3/envs/monai/lib/python3.10/site-packages/torch/jit/_trace.py", line 329, in _check_trace
    copied_dict[name] = _clone_inputs(data)
  File "/homes/clwang/.miniconda3/envs/monai/lib/python3.10/site-packages/torch/jit/_trace.py", line 160, in _clone_inputs
    return function._nested_map(
  File "/homes/clwang/.miniconda3/envs/monai/lib/python3.10/site-packages/torch/autograd/function.py", line 463, in _map
    return fn(obj)
  File "/homes/clwang/.miniconda3/envs/monai/lib/python3.10/site-packages/torch/jit/_trace.py", line 150, in clone_input
    a.detach()
TypeError: MetaTensor.clone() got an unexpected keyword argument 'memory_format'

[KumoLiu]

Hi @ChenglongWang, a workaround can be writer.add_graph(model, tensor(inputs), verbose=True).
Thanks.

[ChenglongWang]

Hi @ChenglongWang, a workaround can be writer.add_graph(model, tensor(inputs), verbose=True).
Thanks.

Thanks! It really helps!