W&B and checkpoints

Logging

Enable Weights & Biases logging:

python src/run_pipeline.py --exp_config configs/XXX.yaml --wandb

Run W&B offline:

python src/run_pipeline.py --exp_config configs/XXX.yaml --wandb --wandb_offline

Loading a model

src/run_pipeline.py can load a trained model:

From W&B:

python src/run_pipeline.py --exp_config configs/XXX.yaml \
  --wandb_load_model wandb \
  --wandb_model_path <entity>/<project>/<run_id>

From a local checkpoint:

python src/run_pipeline.py --exp_config configs/XXX.yaml \
  --wandb_load_model local \
  --local_model_path /path/to/model_best.pt