torch.bernoulli inconsistent gpu/cpu results #10357
Description
Issue description
The behavior of CUDA and CPU sample in the distributions module is inconsistent. On the CPU side, there appears to be an argument validity check, while on the GPU side, this does not appear to be implemented. This issue is somewhat related.
Code example
>>> torch.distributions.Bernoulli(torch.FloatTensor([1.5])).sample()
---------------------------------------------------------------------------
RuntimeError Traceback (most recent call last)
<ipython-input-150-d17a827b2cc3> in <module>()
----> 1 torch.distributions.Bernoulli(torch.FloatTensor([1.5])).sample()
python3.6/site-packages/torch/distributions/bernoulli.py in sample(self, sample_shape)
72 shape = self._extended_shape(sample_shape)
73 with torch.no_grad():
---> 74 return torch.bernoulli(self.probs.expand(shape))
75
76 def log_prob(self, value):
RuntimeError: invalid argument 1: must be >= 0 and <= 1 at /pytorch/aten/src/TH/THRandom.cpp:314
>>> torch.distributions.Bernoulli(torch.cuda.FloatTensor([1.5])).sample()
tensor([1.], device='cuda:0')
>>> torch.distributions.Bernoulli(torch.cuda.FloatTensor([-2])).sample()
tensor([0.], device='cuda:0')System Info
PyTorch version: 0.5.0a0+a24163a
Is debug build: No
CUDA used to build PyTorch: 9.0.176
OS: Ubuntu 16.04.4 LTS
GCC version: (Ubuntu 5.4.0-6ubuntu1~16.04.10) 5.4.0 20160609
CMake version: version 3.9.1
Python version: 3.6
Is CUDA available: Yes
CUDA runtime version: Could not collect
GPU models and configuration:
GPU 0: GeForce GTX 1080 Ti
GPU 1: GeForce GTX 1080 Ti
GPU 2: GeForce GTX 1080 Ti
GPU 3: GeForce GTX 1080 Ti
Nvidia driver version: 390.67
cuDNN version: 7
Versions of relevant libraries:
[pip] numpy (1.13.3)
[pip] numpydoc (0.7.0)
[pip] torch (0.5.0a0+a24163a)
[pip] torchvision (0.1.9)
Metadata
Metadata
Assignees
Labels
No labels