MXNet model serving

Deploy MXNet model serving, with awslab/mxnet-model-server

Dependencies

Python, mxnet, mxnet-model-server, docker, jmeter (for stress tests)

Convert mxnet model (.json & .params) to serving model (.model)

mxnet-model-export --model-name r50 --model-path model-r50-am-lfw

Run standalone mxnet-model-server (flask)

cd models
mxnet-model-server --models r50=r50.model ga=ga.model --service mxnet_vision_service.py --port=8080

Run production mxnet-model-server (docker + nginx + gunicorn + flask)

nvidia-docker run --name mms -p 80:8080 -itd -v <full path to mxnet-serving/models>:/models deepinsight/mms_gpu mxnet-model-server start --mms-config /models/mms_app_gpu.conf

Call mxnet-model-server

curl -X POST http://127.0.0.1/r50/predict -F "data=@models/image.jpg"

Stress test

cd stress-test/threads-100-gpu
jmeter -n -t test-plan.jmx -e -l log -o output

References

https://github.com/awslabs/mxnet-model-server