-
Notifications
You must be signed in to change notification settings - Fork 59
Open
Description
Hi there, I am testing PyTorchWavelets with some minimal code snippets. Unfortunately I encountered the problem with the very first example :-(
# filename is PyTorchWavelets_minitest.py
import torch
import numpy as np
from wavelets_pytorch.transform import WaveletTransform # SciPy version
from wavelets_pytorch.transform import WaveletTransformTorch # PyTorch version
def test():
dt = 0.1 # sampling frequency
dj = 0.125 # scale distribution parameter
batchsize = 32 #! this is where the problem comes. If the problem wont reproduce try increasing this number
# Batch of signals to process
batch = np.random.randn(batchsize,10000)
wa_torch = WaveletTransformTorch(dt, dj, cuda=True)
cwt_torch = wa_torch.cwt(batch)
print(cwt_torch)
if __name__ == "__main__":
test()
It will either give me CUDA error: unspecified launch failure
:
(base) xzc@HP-XZC D:\dev_workspace
> python PyTorchWavelets_minitest.py
Traceback (most recent call last):
File "PyTorchWavelets_minitest.py", line 20, in <module>
test()
File "PyTorchWavelets_minitest.py", line 15, in test
cwt_torch = wa_torch.cwt(batch)
File "d:\Anaconda3\lib\site-packages\wavelets_pytorch-0.1-py3.6.egg\wavelets_pytorch\transform.py", line 310, in cwt
RuntimeError: CUDA error: unspecified launch failure
Or CUDA error: the launch timed out and was terminated
(base) xzc@HP-XZC D:\dev_workspace
> python PyTorchWavelets_minitest.py
Traceback (most recent call last):
File "PyTorchWavelets_minitest.py", line 20, in <module>
test()
File "PyTorchWavelets_minitest.py", line 15, in test
cwt_torch = wa_torch.cwt(batch)
File "d:\Anaconda3\lib\site-packages\wavelets_pytorch-0.1-py3.6.egg\wavelets_pytorch\transform.py", line 310, in cwt
RuntimeError: CUDA error: the launch timed out and was terminated
I am using pytorch 1.0.0, Python version 3.6.5 (both installed through anaconda) on a Windows 10 x64 host. I googled the words "CUDA error: the launch timed out and was terminated" and found this. Not sure if it is corresponding to what is described in that post.
Metadata
Metadata
Assignees
Labels
No labels