Python3 not configured out of the box.

Thank you for your new support for Python3.
The 18.01 image supports tensorflow 1.4.0, but I was looking forward to get the image of tensorflow 1.3.0 with Python3.
Is there a plan to release images supporting for earlier version of tensorflow with Python3? (only tensorflow 1.4.0 and later?)

Great - glad we’re able to help by delivering the requested Python 3 support. The feedback in this thread was very helpful!

Could you share more details around why you’re seeking TensorFlow 1.3.0 rather than 1.4.0? What challenges will 1.3.0 help solve for you that 1.4.0 doesn’t solve?

For context: we’ve been providing the latest versions in the containers each month, given the speed at which the industry is moving, that’s been the request from customers. It’s helpful to receive this request and we’ll need more details to better understand the use cases that’d be solved by providing older versions of a framework.

Since you were asking for feedback, any update on Python 3.6 plans?
Python 3.6 has been the only supported Python 3 version for Amazon Deep Learning AMIs since they launched.
Since I’ve developed/tested on Amazon’s Deep Learning AMI’s, it’s useful to have NGC container that’s compatible.
Also Python 3.6 supports f-strings which is the second best thing since sliced bread (PEP 498 – Literal String Interpolation | peps.python.org)

Great - your comment for Python 3.6 support sparked some internal discussions. The feedback is always welcome and appreciated!

Could you (and anybody else interested) share some more details on how critical having Python 3.6 support is in the NGC container? Does it block your usage of the NGC container? Do you work around it today? If you work around it, what approach do you take?

Do you have specific use cases that you could share which rely on Python 3.6 and won’t work with our Python 3.5 NGC containers?

We all agree with the general direction and the additional details you (and anybody else) can share will be helpful for us to sort out timelines and priorities.

Python 3.6 support is critical for us and without it we are blocked from using the NGC container. Our codebase has been written exclusively for 3.6, with the most prevalent version-specific feature we use being the f-string support.

Hi,

Just wanted to add that I would also highly appreciete if you release ngc images with support for python 3.6

Is there yet any schedule for the release of python 3.6 NGC images?

I encountered the same problem. You can import the package ww (pip install ww). This package has a workaround to make fstrings function in 3.5.

But it would be better to have a 3.6 container from nvidia.