-
Notifications
You must be signed in to change notification settings - Fork 6.8k
updating installation info to have latest packages and more clarity #11503
Conversation
@nswamy - updated Ubuntu instructions as discussed (noting that >16.04 doesn't work). |
docs/install/ubuntu_setup.md
Outdated
@@ -1,67 +1,103 @@ | |||
# Installing MXNet on Ubuntu | |||
|
|||
**NOTE:** For MXNet with Python installation, please refer to the [new install guide](http://mxnet.io/install/index.html). | |||
The following installation instructions are for installing MXNet on computers running **Ubuntu 12 to Ubuntu 16.04**. Support for later versions of Ubuntu is [not yet available](#contributions). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Just to be clear: We only validate ubuntu16.04 - nothing else.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
So should I remove that Ubuntu 12 part?
* [pip Packages](#pip-package-availability) | ||
* [Standard Installation](#standard-installation) | ||
* [Installing Language Packages](#installing-language-packages-for-mxnet) | ||
* [R](#install-the-mxnet-package-for-r) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Is the CPP package missing?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
You're right... this needs to be added to this page.
|
||
Unzip the file and change to the cudnn root directory. Move the header and libraries to your local CUDA Toolkit folder: | ||
Then download [cuDNN 7.1.4](https://developer.nvidia.com/cudnn). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Do we want to specify a minor cuDNN versions? We're always using the latest one in CI, so we shouldn't give the users the impression that only a certain one is supported
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I debated that, but I think if we're recommending an install path, then it should be the one we're testing all of the time. However, we should supply a table of the supported versions per release. I don't know what minor or major versions work for 1.1.0, 1.0.0, and so on....
sudo apt-get upgrade | ||
sudo apt-get install r-base r-base-dev | ||
```bash | ||
wget https://raw.githubusercontent.com/apache/incubator-mxnet/master/ci/docker/install/ubuntu_core.sh |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Puh, I'd prefer if we keep these scripts internal and not user-facing. We install a bunch of crap in there because it's required for some rare use-case. Also, they might be changed at any time. It would be better if we keep these two concerns separate and let the installation guide test ensure that everything works properly.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I was thinking about maintainability. We essentially test these constantly. But, I can agree with the bloat issue. What if we get a stripped down version just for installation instructions, and add that to CI?
Then I can work towards removing the dependency on the install page being parsed for a CI test - something that can end up being brittle and doesn't really thoroughly test as it's pretty much just doing pip installs.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Sure, feel free to create a new nightly test. You could then reference the resources of that test in the installation instructions.
|
||
#### pip Package Availability | ||
|
||
The following table presents the pip packages that are recommended for each version of MXNet. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That's awesome!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I was able to do this because the info is available, but scattered. I wish we had this for the other things like cuDNN and numpy and OpenCV. I notice installation problems with the latter two all of the time.
We should definitely address the >16.04 issues. Ubuntu is our main platform and we want to support it as early as possible to catch early adopters. @nswamy do we have an issue for that so we can look into the problems independent of this PR? |
@marcoabreu - if there's nothing wrong in this info, then we should get it merged ASAP. Then make another issue/jira that captures the things that should be added (like CPP, different install script, addressing >16.04). The site currently shows incorrect info, and I'd like to see that updated right away. |
Okay, I'll merge it due to urgency. Please address it in a new PR. |
…pache#11503) * updating installation info to have latest packages and more clarity * fix table * using image for the table * removed ubuntu 12 mention
Description
There hasn't been an update on installation instructions for Ubuntu in a while. Many new packages are out and not referenced, so I'm trying to fix that.
Comments
pip install mxnet-cu92mkl
is recommended for inference.pip install mxnet-cu92
is recommended for training.