ModuleNotFoundError: No module named 'scaleoutbridge'.

[winca]

When I tried "training_results_v3.1/NVIDIA/benchmarks/resnet/implementations/mxnet", I run cmdline [ CONT="mlperf-nvidia:image_classification-mxnet" DATADIR=/test/training_results_v3.1/NVIDIA/benchmarks/resnet/implementations/run_dir/dataset/preprocessed_data bash ./run_with_docker.s], I just encountered the error: “ModuleNotFoundError: No module named 'scaleoutbridge'”. where can I find the 'scleoutbridge' module?

[Abatpool]

Hi @winca @nv-rborkar

There two .py files in common .

1. Optimizer.py
2. Fit.py.
   We found the respective change for a library scaleoutbridge inside
   /unet3d/implementations/mxnet/runtime/training.py

The changes made are as follows
In optimizer.py we commended out the earlier scaleoutbridge imports and inplace changed it with:-
from mlperf_common.scaleoutbridge import ScaleoutBridgeBase as SBridge
This below is refernce of that change:_

In fit.py
We commented out the native scaleoutbridge and changed it following:-

from mlperf_common.scaleoutbridge import init_bridge,ScaleoutBridgeBase as SBridge
from mlperf_common.frameworks.mxnet import MXNetProfilerHandler, MPICommunicationHandler
Here is the reference image of that change in fit.py:-

Then we have change the sbridge initialization also as follows:-

sbridge = init_bridge(MXNetProfilerHandler(), MPICommunicationHandler(), mllogger)
Here is reference image of the change:-

Alternative way /another way there is another way to bypass the scaleoutbridge error.

There is a scaleoutbridge.py is at the location training_results_v3.1/Fujitsu/benchmarks/resnet/implementations/mxnet
/scaleoutbridge.py in github of mlperf 3.1

Here is the url to find that :-
(

training_results_v3.1/Fujitsu/benchmarks/resnet/implementations/mxnet/scaleoutbridge.py

Line 40 in 5b62935

class ScaleoutBridge(object):

)

Creating a scaleoutbridge.py file at the location
/home/adminserc/training_results_v3.1/NVIDIA/benchmarks/resnet/implementations/mxnet
Then the above changes we had made of changing the imports and Sbridge initialization is not required. The existing code base will take care of it.

[mmarcinkiewicz]

Thanks @Abatpool !

@winca does that fix your issue?

[mmarcinkiewicz]

Alternatively, you can just grab the missing file from here https://github.com/mlcommons/training_results_v4.0/blob/main/NVIDIA/benchmarks/resnet/implementations/eos_n8_ngc24.04_mxnet/scaleoutbridge.py