Hi, the newest tf-nightly-gpu python packages are only built against python 2.7, 3.3, 3.4, 3.5 but not 3.6. My code is not compatible with python 2.
Could you provide an environment option that has python 3.5 installed?
You can use tensorflow-1.2 but the problem is about the CUDA version which should be CUDA 7. I think that the tf-nightly-gpu it's prebuilt with CUDA 9... nvm, I think that it's too much work for you. Which feature of the nightly build do you need?
I have updated the title / category to reflect that this is a new feature request. This will help us gauge interest from our users.
It needs to have CUDA 9 installed indeed. I'm using multiple nightly features, like OneDeviceStrategy, new tf.data features, etc.