A beginner-friendly tool for training and testing PyTorch models without the boilerplate.

Welcome to simple-ml-runner! This project helps students and researchers focus on model design and analysis rather than repetitive training loops.

Simple ML Runner automates the machine learning lifecycle. You define your experiment in a YAML configuration file, and the runner handles:

• Data Loading: Dataset downloading and preprocessing.
• Training: Epoch loops, backpropagation, and optimization.
• Evaluation: Calculating metrics like BLEU or Perplexity.
• Artifacts: Saving checkpoints, logs, and plots automatically.

1. Create a Virtual Environment:

   python -m venv .venv
   # Windows: .venv\Scripts\activate | Unix: source .venv/bin/activate

2. Install Dependencies:

   pip install -r requirements.txt

   Note: For GPU support, follow the PyTorch installation guide.

Run a training job using a configuration file:

ml-runner run -c examples/MNIST/mnist_mlp.yml

Export plots for an experiment:

ml-runner export plot -c examples/nlp/gru_mt_onehot.yml

Override parameters from the command line:

ml-runner run -c examples/MNIST/mnist_mlp.yml --lr 0.01 --epochs 10

For more detailed guides and NLP tutorials, see the docs/ folder:

• NLP Examples Guide
• Project Architecture