[Structured Output][Reasoning] Improves decoding throughput for models using single-token reasoning endings.#30056

chaunceyjiang merged 9 commits intovllm-project:mainvllm-project/vllm:mainfrom

hdlj-h:add-decode-step-end-reasoninghdlj-h/vllm:add-decode-step-end-reasoningCopy head branch name to clipboard