Smart home device control is a difficult task if the instruction is abstract and the planner needs to adjust dynamic home configurations. With the increasing capability of Large Language Model (LLM), they have become the customary model for zero-shot planning tasks similar to smart home device control. Although cloud supported huge LLM models can seamlessly do device control tasks, on-device deployable small LLM shows limited capabilities. In this work, we show how we can leverage huge LLM to enable small LLM for device control task. Towards that goal, we develop a system to generate instruction-plan pairs using huge LLM in an automated manner and used the generated data to finetune small LLMs. We empirically validate the improvement of small LLMs performance on device control task.

https://ieeexplore.ieee.org/document/10446064