LoraConfig

class verta.finetune.LoraConfig(alpha: int = 32, dropout: float = 0.0, r: int = 8)

LoRA fine-tuning configuration.

For use with RegisteredModelVersion.finetune()

Parameters:

• alpha (positive int, default 32) – Scaling factor for update matrices.

• dropout (float between 0.0 and 1.0 inclusive, default 0.0) – Dropout probability for LoRA layers.

• r (positive int, default 8) – Rank of update matrices.