No registered 'RngReadAndSkip', tensor flow-metal 0.3.0

Machine Learning & AI General
miladsa OP
Created Dec ’21
Replies 4
Boosts 0
Views 2k
Participants 4

Hey,

I got my new MBP M1 Max and started setting up the machine for tensorflow. I installed tensorflow-metal as instructed here: https://developer.apple.com/metal/tensorflow-plugin/

Here are some details:

tensorflow-deps           2.7.0                         0    apple

tensorflow-estimator      2.7.0                    pypi_0    pypi

tensorflow-macos          2.7.0                    pypi_0    pypi

tensorflow-metal          0.3.0                    pypi_0    pypi

tensorflow-probability    0.7                        py_1    conda-forge

When instantiating a keras subclassed model, I got the following error:

NotFoundError: No registered 'RngReadAndSkip' OpKernel for 'GPU' devices compatible with node {{node RngReadAndSkip}} . Registered: device='CPU' [Op:RngReadAndSkip]

The code works fine on my new machine when I uninstall tensorflow-metal.

Looking forward to any probable solution. Rooting for running some TF projects on GPU!

Replies  4
Boosts  0
Views  2k
Participants  4
Frameworks Engineer OP
Apple
Jan ’22

Hi @miladsa,

Thanks for reporting the issue. This is being worked on as we are missing the GPU implementation of RngReadAndSkip which is the cause for this error. I will update here once we have added the op to the tensorflow-metal.

0
FatihNurcin OP
Jan ’22

Could you please provide an estimation for next update

1
Frameworks Engineer OP
Apple
Aug ’22

RngReadAndSkip is registered on the GPU in tensorflow-metal==0.5.1.

0
samzzz OP
Jan ’23

RngReadAndSkip isn't registered in tensorflow-metal==0.7.0

What are the differences between 0.7.0 and 0.5.1, and should I downgrade to 0.5.1?

0
No registered 'RngReadAndSkip', tensor flow-metal 0.3.0
First post date Last post date
 
 
Q