Not implemented for GPU

MXNet Model Server
1

Thought I’d give Gluon a whirl, downloaded it, started up the Jupyter, and part way through the “tutorial”, it get this:

MXNetError: [10:37:33] src/imperative/imperative.cc:78: Operator _ones is not implemented for GPU.

Came from executing this:

z = nd.ones(shape=(3, 3), ctx=mx.gpu(0))

I just re-installed mxnet, and same results. Seems odd to me, that a tool written expressly for use with GPUs, is not implemented for GPU usage.

Any suggestions??

Moldy01

2

This looks like a Gluon question. I see that you opened this on “discuss.mxnet.io” as well, which is probably the right channel for this question.

3

Well, my first foray into this world, so I’m not really sure, but from the Error msg, it appeared that it came directly from mxnet, and not Gluon. I can remove the one in Gluon, if you think it is a better fit in the server realm. That’s why I’m asking in both places, get soem thoughts from people with better experience than me.

Thanks for the comments…

4

Did you install the GPU version of MXNet? ‘pip install mxnet’ will give you the CPU version. To get the GPU version, use pip install mxnet-cu90 --pre.

5

OK, so not only did I re-install using the mxnet-cu90 --pre, but before that I totally uninstalled it and tried if from scratch. Then, I downloaded the source from from git, and tried to compile it. There is a error in the code that doesn’t permit the building of the cudnn_algoreg.o, consequently the build fails as well.
I’m guessing that the world just doesn’t want me to play in the mxnet sandbox.

6

I have the same problem. If you or someone have already some solution it would be awesome.