Abstract: Recent advances have demonstrated the potential of decoder-only large language models (LLMs) for automatic speech recognition (ASR). However, enabling streaming recognition within this ...