logger

The recommended loggers are lightning. By default, wandb is configured to store logs locally. If you prefer to use wandb online, manually modify misc.env.WANDB_MODE in tace.yaml.

Note

We do not guarantee that the parameters recorded by Weights & Biases (wandb) are always correct. We recommend directly referring to the errors printed in the standard output. At the end of each epoch, all error metrics are automatically printed.

Example

# Comment out the section below to use Lightning, or uncomment it to use wandb.

# You can specify the logger here, for example, the built-in logger provided
# by Lightning or a third-party logger such as WandB.
logger:
_target_: pytorch_lightning.loggers.WandbLogger
project: ${misc.project_name}
name: ${misc.project_name}
# entity: xuzemin-nanjing-university
log_model: final  # all, final
# tags: ["baseline", "v0.0.1"]
# group: "experiments"
save_dir: wandb_logs