# configs

Python API: `luxonis_train.configs`

Ready-to-use config files shipped with the package.

The `train`, `tune`, `inspect`, `test`, `infer`, `annotate`, `export`, `archive`, `convert`, and `quantize` commands accept
`--model` and an optional `--variant` instead of `--config`. A run then needs no local config file. For example, `luxonis_train
train --model detection --variant light` loads `detection_light_model.yaml`. The Python equivalent is
`LuxonisModel(model="detection", variant="light")`.
[luxonis_train.config.predefined](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/config/predefined.md)
finds the file.

`--variant` accepts every variant that the predefined model declares, not only the variants with a file of their own. For example,
`--model detection --variant medium` loads `detection_light_model.yaml` and selects the `medium` variant, because no
`detection_medium_model.yaml` exists. Run `luxonis_train list-models` to see the model names and their variants.

The other files are examples:

 * `defaults.yaml` shows the default values of the config fields. The package does not load it.
 * `complex_model.yaml` builds a node graph by hand instead of naming a predefined model.
 * `example_export.yaml` and `example_tuning.yaml` show the `exporter` and the `tuner` sections.