Closed 00krishna closed 5 years ago
It looks like Cudnn and cudatoolkit were both updated recently. Date: Mon Jun 3 2019: cudnn 7.6.0 for cuda 9.0, 10.0, 10.1 & cudatoolkit 10.1.168
Tensorflow 2.0 will not be available from Anaconda while it is tagged as pre-release (alpha/beta). Tensorflow is still listed as being pre-release here: https://github.com/tensorflow/tensorflow/releases
@csoja thanks for the info. I appreciate your post. Yes, I just happened to do an update on conda
environment and saw that the cudnn
version went to 7.6.0 and the cudatoolkit
version was also upgraded. So that is good. I had found some hacky way around this before, but now that the regular conda
environment is updated I can use that.
Was there any kind of announcement about this update? I did not see any tweet or message on the README about this. I mean I posted this issue 10 days ago, so the Anaconda folks must have already been testing the updated cudnn
and cudatoolkit
versions when that message was posted. I also posted this issue at https://github.com/conda/conda/issues/8736 because I was not sure the appropriate location to post the issue.
I think the only message I received on this issue was that Anaconda does not support unreleased software and Tensorflow-2.0 alpha was not released. I know that you were not personally on that issue chain, but hopefully you can understand my own frustration a bit. I mean, the commentor could have just said that "we are testing updated cudnn and cudatoolkit and it will be released in 5 days."
Don't worry I am not trying to rant or anything. But I think that the Anaconda teams still have really big communications issues with the users. I am sorry to say that, but it really does affect us trying to use Anaconda.
Once again, I appreciate your getting back to me.
@00krishna Are you aware of the RSS feed? https://repo.anaconda.com/pkgs/rss.xml This lists the most recent package updates to repo.anaconda.com This is the current method for announcing package updates. There are too many to tweet about every update. Also, if you want to browse what is available in the repositories, you can do so from here: https://repo.anaconda.com/pkgs/
It is a very small team and they do the best they can to keep up, but with millions of users and many, many public repositories and mailing lists they are responsible for - it is an uphill battle. Thanks for your patience and for using Anaconda.
@csoja I really appreciate your getting back to me. I did not know about the RSS feed, so I will keep any eye on that in the future. I know that you guys are working hard and it is quite amazing how well the Anaconda environment works considering the number of packages that are integrated into the ecosystem. I remember the challenges of the previous Enthought ecosystem, and then before that trying to install and compile all of the packages myself, haha. So glad that those days are behind us.
Keep up the good work. And definitely anything you can do to help we users to understand the Anaconda plans for the future, or dealing with different challenges, etc., is really helpful. This Tensorflow 2.0 thing was a really great example where Google is pushing everyone to start using Tensorflow 2.0. All of the tutorials and communications are geared towards users migrating early so that they can avoid a last minute scramble when Tensorflow 2.0 is finally an official release. So in parallel, communications from Anaconda/ContinuuumIO are helpful to understand how Anaconda will respond to the challenge posed by TF 2.0.
But I think that this is just an example of upcoming challenges to the ecosystem that are on the horizon. I think Python 2.7 will End-of-Life on Jan 1, 2020. Anaconda has already been communicating about that transition and that has been an example of very proactive communications. I think those communications prompted me to retire some docker containers that were still lingering in my repos, etc.
Thanks again and best of luck.
Actual Behavior
My basic question is whether there is a date set for upgrading from Cudnn 7.3.1 to a newer version. Also, when will Anaconda move to a newer version of the cudatoolkit?
Most Tensorflow 2.0 stuff works in Anaconda. But I ran into one particular issue where I got an error like this.
I opened an issue on the Conda repo, but the issue was closed with essentially no response--other than saying that Anaconda does not support Tensorflow 2.0
https://github.com/conda/conda/issues/8736
So I was just wondering when updated libraries would be available?
Expected Behavior
Steps to Reproduce
I created a conda environment with
conda create -n tf2alpha tensorflow-gpu==2.0.0-alpha0 anaconda
Then I created a notebook and ran the code below.
Anaconda or Miniconda version:
Operating System:
conda info
conda list --show-channel-urls