🐛 Bug
I am adding a new operator tensor.view(dtype) in #47951, and getting lots of JIT failures.
With #47951, the script to reproduce this issue is
import torch
def f(x):
return x.view(-1)
s = torch.jit.script(f)
print(s.graph)
x = torch.randn(5, 5, 5)
s(x)and there will be an error message
graph(%x.1 : Tensor):
%4 : int = prim::Constant[value=-1]() # jit-view.py:4:18
%5 : Tensor = aten::view(%x.1, %4) # jit-view.py:4:11
return (%5)
Traceback (most recent call last):
File "jit-view.py", line 10, in <module>
s(x)
RuntimeError: The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
File "jit-view.py", line 4, in f
def f(x):
return x.view(-1)
~~~~~~ <--- HERE
RuntimeError: Viewing a tensor as a new dtype with different bytes is not supported.
cc @gmagogsfm @mruberry
🐛 Bug
I am adding a new operator
tensor.view(dtype)in #47951, and getting lots of JIT failures.With #47951, the script to reproduce this issue is
and there will be an error message
cc @gmagogsfm @mruberry