callbacks#

A callback refers to a set of functions that are automatically invoked after the completion of each training epoch. At a minimum, you need at least one callback for saving the model. If you have any special requirements, or if you want to use any built-in features provided by Lightning, you can add them here as callbacks.

Note

Below is an example of usage. If a parameter comes from the internal implementation of TACE, it may not be the most up-to-date. For the latest parameters, please refer to the corresponding configuration files on GitHub. A complete list of parameters along with detailed explanations is provided.

Example#

callbacks:
    early_stopping:
        _target_: lightning.pytorch.callbacks.EarlyStopping
        verbose: true
        log_rank_zero_only: true
        monitor: ${synth_metric.monitor_metric_name}
        min_delta: 1e-5
        patience: 80

    ema: # ema is always recommended
        _target_: tace.utils.callbacks.EMACallback
        decay: 0.995 # typically >= 0.99
        use_num_updates: true

    checkpoint: # at leas one checkpoint is required
        _target_: lightning.pytorch.callbacks.ModelCheckpoint
        dirpath: checkpoints
        filename: TACE-OMat24-{epoch}-{step}-{${synth_metric.monitor_metric_name}:.4f}
        # monitor: ${synth_metric.monitor_metric_name}
        save_top_k: -1
        save_last: true
        every_n_epochs: 1
        # every_n_train_steps: 2000
        # mode: min
        save_weights_only: false
        auto_insert_metric_name: false
        verbose: true

# checkpoint_step:
#   _target_: lightning.pytorch.callbacks.ModelCheckpoint
#   dirpath: checkpoints
#   filename: TACE-OMat24-{epoch}-{step}-{${synth_metric.monitor_metric_name}:.4f}
#   # monitor: ${synth_metric.monitor_metric_name}
#   save_top_k: -1
#   # every_n_epochs: 1
#   every_n_train_steps: 2000
#   mode: min
#   save_weights_only: false
#   auto_insert_metric_name: false
On this page