Cntk: Can MSFT make a version of CNTK for MacOS Sierra 10.12?

Created on 2 Jun 2017  路  17Comments  路  Source: microsoft/CNTK

Like their version of VSCode for Mac, could they follow with CNTK prebuilt pip install for MacOS?

Thanks!

feature request

Most helpful comment

@cha-zhang I care deeply the wide adoption of CNTK to all platforms AND NOT SIMPLY to be as good OR exceed tensorflow.

Not having budget for Mac build in my view is MISSING the opportunity [ The same way Microsoft MISSES the dawn of iPhone]

There is some Cross disciplinary feedback on CNTK.

Do consider establishing WIN WIN inter-departmental collaboration with guys from XAMRIN to make CNTK truly the LEADER in cross platform [through XAML Standard and Xamarin Forms, to iOS, MacOS, Android, Linux GTK, WPF and UWP] deep learning library.

All 17 comments

We can in theory, but we currently don't have the infrastructure to test Mac build. If you have any ideas, let us know.

@cha-zhang I care deeply the wide adoption of CNTK to all platforms AND NOT SIMPLY to be as good OR exceed tensorflow.

Not having budget for Mac build in my view is MISSING the opportunity [ The same way Microsoft MISSES the dawn of iPhone]

There is some Cross disciplinary feedback on CNTK.

Do consider establishing WIN WIN inter-departmental collaboration with guys from XAMRIN to make CNTK truly the LEADER in cross platform [through XAML Standard and Xamarin Forms, to iOS, MacOS, Android, Linux GTK, WPF and UWP] deep learning library.

Thanks for the suggestion, @JimSEOW! Will take your advice into serious consideration when planning for future CNTK releases.

It will be great to support macOS and help spread CNTK because a huge number of developers are using macOS. Looking forward for the support!

@cha-zhang @cha-zhang
Both (MacOS) iOS and Android platforms are critical for CNTK mobile strategy especially mobile devices and the latest Microsoft Always connected cellular PC (Q4 2017) will be powered by proven battery efficient Snapdragon 835 that has been recently shown to be on par with mid range intel/AMD CPU.

For AI (Deep Learning), the more time it can learn (through the always connected mobile PC, the smarter will the intelligent solution.

It is critical to the future of Mobile AI that it is there is a platform (CNTK) that mobilize _mobile programmers from iOS and Android_ and Windows on ARM - all using .NET CORE Standard and perhaps XAML standards that are available in so many OS as listed above.

=> in other words, in one to two years, if you do not program AI in both mobile and server platforms, your career is over :-)

@cha-zhang What type of "infrastructure" is holding you back? You, MSFT, use Macintoshes in your presentations for Channel9 and make VSCode for Mac along with Visual Studio for Mac? Why not increase the usage of CNTK because I have yet to see any papers besides MSFT that use CNTK for the deep learning experiments AFAIK?

@shyamalschandra We don't have infrastructure to build and test Mac version of CNTK regularly. I don't mean we can't, I mean we haven't got the resource to work on that. It's a gap the team is very eager to fill.

Usage of CNTK inside MSFT is high, but externally most people only use CNTK for speech research. We are going to CVPR this year to give a CNTK tutorial, and we hope to attract more vision researchers to try CNTK.

Hello,

Instead of a full mac port, how about getting the evaluation library working on Mac first? I am using CNTK extensively ( since version 1.7 ) and it's fantastic. I deploy it on native windows apps but will love to deploy the prediction functionality on my mac versions as well.

This will make CNTK much more prevalent on other platforms as well!

Cheers

if you take a note out of tensorflow's book - a C-API or c++ API would help foster development across platforms.
https://github.com/tensorflow/tensorflow/issues/19

Tensorflow has no plans to do training support beyond python.
I think if you can expose full API / then developers would rally to contribute to extending platform.
If you can get Node(as this seems to be getting most attention on tensorflow ports) to interact with CNTK - this would be pretty compelling.

did I mention protobuffers?

Our C++ API support everything supported by our Python API. CNTK Python API is build on the top of our publically available C++ API. This is why most of our code is in C++.

We use protobuffer for serialization.

@cha-zhang I think I can better appreciate what you're saying about an "infrastructure" for testing Macs because CICD scripts need a "farm" of machines to run unattended. Well, there is an outfit -
https://www.macstadium.com

Supporting Mac OS is on or roadmap. It might take a few months, though :)

@cha-zhang Hey I hope this doesn't come across the wrong way but it came up internally because a customer brought it to our attention.
I don't understand enough about AI as I probably should by any means but am intrigued by your project and would like to help you out and allowing more developers (like me) to get started more easily. If you need help with Mac infrastructure I'm sure that we can help out one way or another.

We're proud to have recently started offering a F/OSS plan for you to test with and see where it leads you.

Is there any progress on this front?

@mnboos: Well, I have tried this technique with other Linux-only repositories: 1.) Replace all the dependencies from apt-get into equivalent brew commands inside the make or cmake configuration file. 2.) Invent your own clones of Linux dependencies if they are not there. 3.) compile from source. 4.) Enjoy the CNTK!

Any progress on this? I would love to have a pip install of CNTK.

AFAIK, nothing. I can work on it as soon as I have time.

Was this page helpful?
0 / 5 - 0 ratings

Related issues

arijit17 picture arijit17  路  3Comments

IvanFarkas picture IvanFarkas  路  4Comments

SudharakaP picture SudharakaP  路  5Comments

christopher5106 picture christopher5106  路  5Comments

springkim picture springkim  路  4Comments