rlan/ray-pytorch-lightning-mlflow

Example MNIST training using the following MLops tech stack: Ray, PyTorch Lightning, PyTorch, MLflow and Python 3.10.12.

READ ME

Example MNIST training using the following tech stack:

• Python 3.10.12
• Ray
• PyTorch
• PyTorch Lightning
• MLflow

Files

• mnist.py. Main training. Example from Ray's documentation.
• requirements.txt. Install via: pip install -r requirements.txt
• freeze.txt. Resulting run-time environment.

Fixes

In ray.tune.examples.mnist_ptl_mini package, change

self.accuracy = Accuracy()

to

self.accuracy = Accuracy(task="multiclass", num_classes=10)

Run

python mnist.py

The experiments will be recorded to MLflow in a local folder, mlruns and to Ray Tune in ~/ray_results/tune_mnist.

Launch MLflow in the launch folder.

mlflow ui