I am trying to install pytorch_ard to test variational dropout, but whatever the install of the library, the test script constantly fails with the same error:
Sparsification ratio: 11.880%
Traceback (most recent call last):
...
/pytorch_ard-0.2.1-py3.6.egg/torch_ard/torch_ard.py", line 70, in _update_sparse_weights
File "/home/supercalculateur/environment/torch_ard/lib/python3.6/site-packages/torch/tensor.py", line 325, in rsub
return _C._VariableFunctions.rsub(self, other)
RuntimeError: Subtraction, the - operator, with a bool tensor is not supported. If you are trying to invert a mask, use the ~ or bitwise_not() operator instead.
I tried :
to install everything with pypi
to docwngrade torch to 1.1
to install pytorch_ard, from source (with gcc 5.5 instead of standard 7.4)
It must be silly, but I still can't figure out why I always get this error (either operation is not defined or tensor is wrongly a boolean one)
I use nvidia driver 430.26 with CUDA Version: 10.2 on ubuntu 18.04
Hi,
I am trying to install pytorch_ard to test variational dropout, but whatever the install of the library, the test script constantly fails with the same error:
Sparsification ratio: 11.880% Traceback (most recent call last): ... /pytorch_ard-0.2.1-py3.6.egg/torch_ard/torch_ard.py", line 70, in _update_sparse_weights File "/home/supercalculateur/environment/torch_ard/lib/python3.6/site-packages/torch/tensor.py", line 325, in rsub return _C._VariableFunctions.rsub(self, other) RuntimeError: Subtraction, the
-
operator, with a bool tensor is not supported. If you are trying to invert a mask, use the~
orbitwise_not()
operator instead.I tried :
It must be silly, but I still can't figure out why I always get this error (either operation is not defined or tensor is wrongly a boolean one)
I use nvidia driver 430.26 with CUDA Version: 10.2 on ubuntu 18.04
Regards,
Francois