Troubleshooting

Configuration path issues

• Ensure train.py uses names from config/model/ without the .yaml extension.

• Ensure plugin names map to a file in config/plugin/.

Runtime issues

• If GPU setup is unavailable, switch to CPU in config:

accelerator: "cpu"
gpu_devices: []

• If out-of-memory occurs, lower train_batch_size.

• If training is unstable, reduce learning rate or disable mixed precision.

Tracking issues

• Set wandb: False if Weights and Biases tracking is not desired.

Reproducibility suggestions

• Use fixed seeds.

• Keep config files versioned with experiment logs.

• Record the git commit hash with each run.