AutoTimmCLI¶

YAML-config-driven command-line interface built on LightningCLI.

AutoTimmCLI¶

A subclass of LightningCLI that uses AutoTrainer by default and discovers all AutoTimm task and data module classes automatically.

API Reference¶

36
37
38
39
40
41
42
43
44
45
46
47

class AutoTimmCLI(LightningCLI):
    """AutoTimm command-line interface built on LightningCLI.

    Supports subcommands: ``fit``, ``validate``, ``test``, ``predict``.
    Uses ``AutoTrainer`` as the default trainer class and discovers all
    AutoTimm task and data module classes automatically.

    Example:
        >>> # From the command line:
        >>> # autotimm fit --config config.yaml
        >>> # autotimm fit --config config.yaml --trainer.max_epochs 20
    """

Usage¶

# Train with a YAML config
autotimm fit --config config.yaml

# Or via python module
python -m autotimm fit --config config.yaml

# Override parameters
autotimm fit --config config.yaml --trainer.max_epochs 20

# Show help
autotimm --help
autotimm fit --help

Subcommands¶

main()¶

Entry point function for the CLI.

main() -> None

50
51
52
53
54
55
56
57
58
59
60

def main() -> None:
    """Entry point for the AutoTimm CLI."""
    AutoTimmCLI(
        model_class=pl.LightningModule,
        datamodule_class=pl.LightningDataModule,
        trainer_class=AutoTrainer,
        subclass_mode_model=True,
        subclass_mode_data=True,
        save_config_callback=None,
        seed_everything_default=False,
    )

Usage¶

Called automatically when using the autotimm command or python -m autotimm.

Can also be called programmatically:

from autotimm.cli import main

main()

Config File Format¶

The CLI expects a YAML config with model, data, and trainer sections:

model:
  class_path: autotimm.ImageClassifier
  init_args:
    backbone: resnet18
    num_classes: 10

data:
  class_path: autotimm.ImageDataModule
  init_args:
    dataset_name: CIFAR10
    data_dir: ./data
    batch_size: 32
    image_size: 224

trainer:
  max_epochs: 10
  accelerator: auto

Supported Model Classes¶

Supported Data Modules¶

Standalone CLI Modules¶

In addition to the autotimm CLI, AutoTimm provides standalone CLI modules for specific tasks:

export_jit — TorchScript Export¶

Export a trained checkpoint to TorchScript (JIT) format:

python -m autotimm.export.export_jit \
    --checkpoint path/to/checkpoint.ckpt \
    --output model.pt \
    --task-class ImageClassifier \
    --input-size 224

See also: Export API Reference

interpret_cli — Model Interpretation¶

Run interpretation methods on a trained checkpoint from the command line:

python -m autotimm.cli.interpret_cli \
    --checkpoint path/to/checkpoint.ckpt \
    --image path/to/image.jpg \
    --methods gradcam,gradcampp,integrated_gradients \
    --output-dir ./interpretations \
    --task-class ImageClassifier

Output: JSON to stdout with heatmap file paths, predicted class, and per-method errors.

{
  "results": {"gradcam": "/path/to/gradcam.png", ...},
  "predicted_class": 5,
  "errors": {}
}

Available methods: gradcam, gradcampp, integrated_gradients, smoothgrad, attention_rollout, attention_flow

See also: CLI User Guide | AutoTrainer | Export API | Example Configs