ksuit.initializers.previous_run_initializer

Classes

PreviousRunInitializer

Initializes a model from a checkpoint of a previous run (specified by the run_id), this initializers assumes that the previous run is over, and hence only loads model weights.

Module Contents

class ksuit.initializers.previous_run_initializer.PreviousRunInitializer(initializer_config, **kwargs)

Bases: ksuit.initializers.base.CheckpointInitializer

Initializes a model from a checkpoint of a previous run (specified by the run_id), this initializers assumes that the previous run is over, and hence only loads model weights. When a previous run should be resumed for further training, use ResumeInitializer instead.

Example

```yaml
model:

… # model config dim: 64 initializers:

  • kind: ksuit.initializers.PreviousRunInitializer model_info: ema=0.9999 checkpoint: latest model_name: pointnet run_id: 2025-11-01_8zlau

```

When training, the stage name is usually “train”. Checkpoint is usually “latest” or “best_loss”, depending on which checkpoint you want to load. Model_info is an optional string that is used as suffix in the checkpoint filename. Model_name is the name of the model to load, which should match the model_name used in CheckpointCallback during the previous run. :param initializer_config: Configuration for the initializer. Implements PreviousRunInitializerConfig schema. :param **kwargs: additional arguments to pass to the parent class.

Initialize model from checkpoint.

Parameters:

  • initializer_config (ksuit.schemas.initializers.PreviousRunInitializerConfig) – configuration for the initializer. Implements the CheckpointInitializerConfig schema.

  • **kwargs (dict) – additional arguments to pass to the parent class.

keys_to_remove
patterns_to_remove
patterns_to_rename
patterns_to_instantiate
init_weights(model, model_name=None)

Initialize the model weights from the checkpoint.

Parameters:
Return type:

None