Doc-to-LoRA Uses Hypernetworks to Generate LoRA Adapters On-the-Fly

Sakana AI's Doc-to-LoRA trains a hypernetwork to generate LoRA adapters directly from documents or task descriptions, enabling instant LLM customization without per-task fine-tuning runs. The approach amortizes training cost across tasks and targets the gap between static LLM weights and rapid in-context adaptation.