Closed tm157 closed 7 years ago
As far as I can see it, it is mainly an allocation error. Can you suggest any changes to the code in order to circumvent this issue?
Yeah, that looks like an allocation failure.
Are you doing anything else on the GPU at the moment? If so, that might use up part of the GPU memory and cause there to be not enough room to train. If not, perhaps your GPU is too small to process the default number of batches. To fix that, you can try changing the line
allow_gc=False
to
allow_gc=True
in your ~/.theanorc file to enable garbage collection. If that doesn't work, you can reduce the batch size here. You might also look at this part of the Theano FAQ
I am getting the following error
RuntimeError: CudaNdarray_ZEROS: allocation failed.
is this because of theano version?
Thanks