Researchers at Sakana AI have developed Text-to-LoRA (T2L), a hypernetwork designed to generate task-specific Low-Rank Adaptation (LoRA) adapters for large language models (LLMs) based solely on natural language task descriptions. This approach eliminates the need for separate fine-tuning for each downstream task, enabling instant adaptation of LLMs by producing new LoRAs in a single forward pass. T2L can also compress many existing LoRAs into itself, allowing for efficient and dynamic model adaptation. The technology represents a shift from traditional adapter tuning, which typically requires extensive computational resources and time. The innovation was presented at ICML 2025 and aims to facilitate rapid task switching for LLMs by converting plain English prompts into plug-and-play model upgrades. This development could significantly reduce the cost and complexity of customizing LLMs for diverse applications.