[JIT] invoking nn.Sequential requires all intermediary results to be the same type #27401
Description
🐛 Bug
To Reproduce
class A(torch.nn.Module):
def __init__(self):
super(A, self).__init__()
def forward(self, x):
return x + 3
class B(torch.nn.Module):
def __init__(self):
super(B, self).__init__()
def forward(self, x):
return {"1": x}
class C(torch.nn.Module):
__constants__ = ['foo']
def __init__(self):
super(C, self).__init__()
self.foo = torch.nn.Sequential(A(), B())
def forward(self, x):
return self.foo(x)
c = C()
print(torch.jit.script(c).graph)
RuntimeError:
Variable 'input' previously has type Tensor but is now being assigned to a value of type Dict[str, Tensor] : at :4:16
def forward(self, input):
for m in self:
input = m(input)
~~~~~ <--- HERE
return input
Expected behavior
Either this should work or be documented as not working.
cc @suo