add_graph_onnx not working

Maybe this is something that would be fixed with a new release on pip?

I had the following code:

writer.add_graph(model, images[:2])

I got the following warning:

± python a1_mnist_fashion.py
warning: Embedding dir exists, did you set global_step for add_embedding()?
You are using PyTorch==0.3, use add_graph_onnx()

so I changed my code to:

# add the graph to tensorboard for visualization
writer.add_graph_onnx(model)

and then I got:

± python a1_mnist_fashion.py
warning: Embedding dir exists, did you set global_step for add_embedding()?
Traceback (most recent call last):
  File "a1_mnist_fashion.py", line 129, in <module>
    writer.add_graph_onnx(model)
  File "/home/ahundt/.local/lib/python2.7/site-packages/tensorboardX/writer.py", line 377, in add_graph_onnx
    self.file_writer.add_graph_onnx(gg(prototxt))
  File "/home/ahundt/.local/lib/python2.7/site-packages/tensorboardX/graph_onnx.py", line 11, in gg
    m = onnx.load(fname)
  File "/home/ahundt/.local/lib/python2.7/site-packages/onnx/__init__.py", line 29, in load
    with open(obj, 'rb') as f:
TypeError: coercing to Unicode: need string or buffer, Net found
-> [1]

[amorgun]

@lanpa Sadly add_graph() is not working with 0.3.1 because of signature mismatch:

1. https://github.com/lanpa/tensorboard-pytorch/blob/0c9f7ba28b37e4f22f269e4350e1c5813a3b5634/tensorboardX/graph.py#L45 - two arguments are used
2. https://github.com/pytorch/pytorch/blob/v0.3.1/torch/onnx/__init__.py#L78 - only one argument is allowed Also, that check does not allow to use add_graph() in 0.3.1. EDIT: It was fixed in https://github.com/lanpa/tensorboard-pytorch/pull/83, sorry for bothering you.

[lanpa]

@ahundt please have a test~