tensorboardx当网络模型输入两个参数时add_graph()报错解决方法

pytorch:0.4.0
python:3.6

File "/user/files/apps/anaconda2/envs/pytorch/lib/python3.6/site-packages/tensorboardX/graph.py", line 85, in graph
list_of_nodes = parse(graph)
File "/user/files/apps/anaconda2/envs/pytorch/lib/python3.6/site-packages/tensorboardX/graph.py", line 61, in parse
n['inputs'][i] = mapping[s]
KeyError: '408'

解决方法：

If I change one line in tensorboardx package (graph.py line 61 of the function parse), everything works fine:

Original:
n['inputs'][i] = mapping[s]
New:
n['inputs'][i] = mapping[s] if s in mapping else ""

具体不知道为啥，之后知道原因后会更新