I'm running into the error documented here. The source of the error is a bug in cuDNN, which apparently was fixed in version 7.0.5.
I checked the cuDNN version in both the standard environment and the tensoflow-1.5 environment, and it appears to be 7.0.4. So my questions are:
- Do you support higher versions of cuDNN?
- If not, do you plan do so in the near future?