# optimizers

Python API: `luxonis_train.optimizers`

Optimizers a config can name.

[luxonis_train.optimizers.optimizers](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/optimizers/optimizers.md)
registers twelve optimizers of `torch.optim` in the `OPTIMIZERS` registry, under their class names. `trainer.optimizer.name`
accepts any of them, and `trainer.optimizer.params` goes to its constructor.

[CompositeOptimizer](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/optimizers/composite_optimizer.md)
wraps the inner optimizers of a training plan when the plan has more than one. Lightning then sees one optimizer, so gradient
accumulation and gradient clipping still work.

## Child Pages

 * [composite_optimizer](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/optimizers/composite_optimizer.md)
 * [optimizers](https://docs.luxonis.com/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/optimizers/optimizers.md)