As large language models (LLMs) become more powerful and complex, efficiently running them for real-world applications is increasingly challenging.