Training stops after successfully opening libcudnn.so.7 in colab

user_19806416 0
Viewed 33

I'm using google colab to train fasterrcnn with tensorflow 1.15 installed. With the free GPU, the training stops after "successfully opened libcudnn.so.7". I tried training without GPU as well and the training stops after step 1. Similar problems have been reported before, but I can't seem to find a solution that works. Will really appreciate any advice on this issue.

This is when the training was done without GPU in colab

1

This is when training was done with GPU in colab

2

training google-colaboratory tensorflow
1 Answers

How have you installed Tensorflow 1.15 on Colab? I think that besides reinstalling with pip you also have to downgrade CUDA:

!pip uninstall tensorflow
!pip install tensorflow-gpu==1.15
!apt install --allow-change-held-packages libcudnn7=7.4.1.5-1+cuda10.0

Here a table with compatibility versions. If you want to see all the versions available for a package like libcudnn do:

!apt list -a libcudnn7
user_147749590
Related