If vLLM already solved LLM serving, why did SGLang appear?

After the launch of ChatGPT and open-source models in 2022–2023, lots of companies tried hosting models on GPU infrastructure, but they encountered issues like GPU idle time and computational bottlene