Cannot update ModelDict with another ModelDict

1

Hi. I’m new to this forum, so I hope I’m making myself clear enough :sweat_smile:. I’m working on a project using pytorch 1.4.0 and I’m getting an error while trying to merge two ModelDict instances. According to the documentation of torch.nn.ModelDict, a ModelDict object can be appended to another one using the update() function. I tried this with the following code:

model1 = torch.nn.ModuleDict(OrderedDict([
          ('conv1', torch.nn.Conv2d(1,20,5)),
          ('relu1', torch.nn.ReLU())]))

model2 = torch.nn.ModuleDict(OrderedDict([
          ('conv1', torch.nn.Conv2d(1,20,5)),
          ('relu1', torch.nn.ReLU())]))

model1.update(model2)

but this gives the following error:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-157-967ebf8b86d0> in <module>
     10 print(type(model2))
     11 
---> 12 model1.update(model2)

c:\python36\envs\modern_pointcn\lib\site-packages\torch\nn\modules\container.py in update(self, modules)
    348                     raise ValueError("ModuleDict update sequence element "
    349                                      "#" + str(j) + " has length " + str(len(m)) +
--> 350                                      "; 2 is required")
    351                 self[m[0]] = m[1]
    352 

ValueError: ModuleDict update sequence element #0 has length 5; 2 is required

by looking into the source code, it seems a torch.nn.ModelDict is not recognized as a Mapping. Is this a bug? If not, how can I append two ModelDict instances?

2

This might be a bug, as the condition is apparently checking the length of the key instead of the tuple.
Could you please create an issue on GitHub?

3

Thank you for the feedback. Yes, I’ll open an issue on Github.