Can anybody provide a kv-store (distributed training) example with cpp package?

[xiaolin-cheng]

I am OK with python bindings, but not able to figure out how to use cpp package kv-store APIs. There is no working example under mxnet\cpp-package\example. Thanks a lot!

[kalyc]

Thanks for submitting the issue @xiaolin-cheng
@mxnet-label-bot[Example, C++]

[xiaolin-cheng]

Thanks for submitting the issue @xiaolin-cheng
@mxnet-label-bot[Example, C++]

@kalyc do we have a C++ example of distributed training? thanks

[leleamol]

@xiaolin-cheng We currently do not have example in C++ to demonstrate the distributed training.

We have the support for distributed training in Python. Would you be able to share the use cases where you are planning to use distributed training using C++? I am especially curious about the reasons behind choosing C++ over Python for training.

Thanks

[xiaolin-cheng]

@kalyc @leleamol

I am implementing an allreduce distributed training algorithm, and wanted to compare with MXNet's native parameter server approach. I did some hacking in the library C++ code, so comparing with C++ ps training would be fair. A quick C++ ps example of even how to use C++ ps APIs would be great. I was lost when I looked into C++ ps APIs and had no idea where to start.

Thanks

@xiaolin-cheng We currently do not have example in C++ to demonstrate the distributed training.

We have the support for distributed training in Python. Would you be able to share the use cases where you are planning to use distributed training using C++? I am especially curious about the reasons behind choosing C++ over Python for training.

Thanks

[leleamol]

@xiaolin-cheng Thanks for providing the information regarding your use case.

The C++ API that are meant to be used as front-end API are available [here] (https://github.com/apache/incubator-mxnet/tree/master/cpp-package/include/mxnet-cpp)
There are examples that demonstrate training workflows using simple networks.

Currently, the documentation for these APIs are not getting generated and published. We have a PR #12977 out that will add the documentation to the main site.

[sojiadeshina]

@xiaolin-cheng were you able to implement what you needed with the generated c++ documentation? if you were, you can contribute your code as an example for distributed training with c++

[pinaraws]

@mxnet-label-bot add[Distributed]

[timespaceuniverse]

@xiaolin-cheng , i also want to have a distributed-kvstore c++ example, do you have one now ?
otherwise i have to check the python api call process.