aliabbasjaffri / mnist-zenml-bentoml-pipeline

A pipeline built on ZenML for training digit MNIST dataset using Pytorch, experiment tracking using MLFlow and model deployment using BentoML

mnist-zenml-bentoml-pipeline

This project is based on top of MNIST example project from BentoML.

• This project is dependent on python and runs perfectly on version 3.6, 3.7 and 3.8

• You can always install required python version using pyenv tool.

• After setting up python, install requirements

• Next, setup zenml by running zenml init in the root folder. This creates a .zen folder in the root directory that tracks your progress.

• Inspect the file zenml_pipeline.py. Observe the different steps the pipeline is composed of.

• Run the training pipeline that ends up saving the model to registry using

zenml integration install mlflow
python3 zenml_pipeline.py

• Run the generated model via:

bentoml serve service:svc --reload

• With the --reload flag, the API server will automatically restart when the source file service.py is being edited, to boost your development productivity.
• Verify the endpoint can be accessed locally:

curl -H "Content-Type: multipart/form-data" -F'fileobj=@samples/1.png;type=image/png' http://127.0.0.1:5000/predict_image