A tweak to LoRA fine-tuning splits its capacity into separate orthogonal experts, beating standard LoRA with zero extra inference cost.
Continue to AI University →