Examples

Forge ML ships with small demo models for several frameworks. These examples train quickly on random data and demonstrate how to save the trained weights to a local path or a Minio bucket.

Running Examples

Each framework exposes make_* functions which build and train the models. They accept optional save_path and minio_cfg keyword arguments. If a saved model exists the weights can be reloaded using load=True.

from forge_ml.examples.pytorch import cnn
model = cnn.make_cnn(save_path="cnn.pt")

# Upload to Minio
cnn.make_cnn(save_path="cnn.pt", minio_cfg={
    "endpoint": "localhost:9000",
    "access_key": "minioadmin",
    "secret_key": "minioadmin",
    "bucket": "forge-examples",
})

See the modules under forge_ml/examples for TensorFlow and JAX versions of LSTM, RNN, CNN, FPN, ResNet and FCN models.

Jupyter Notebooks

Interactive notebooks demonstrating these models and the testing pipeline are available under notebooks/examples. Open one of pytorch_demo.ipynb, tensorflow_demo.ipynb or jax_demo.ipynb in JupyterLab to train the models and execute the pipeline tests. A reinforcement learning demo using Farama Gym can be found in reinforcement/cartpole_demo.ipynb.