我有一台服务器是8张64G的K100-AI，运行Qwen3.5-122B-A10B模型，具体部署如下： docker容器启动： qwen3.5: image: harbor.sourcefind.cn:5443/dcu/admin/base/custom:vllm018-ubuntu22.04-dtk26.04-gemma4-0413 container_name: qwen3.5 privileged: true user: root network_mode: host ipc: host shm_size: 1024G devices: - "/dev/kfd:/dev/kfd" - "/dev/mkfd:/dev/mkfd" - "/dev/dri:/dev/dri" volumes: - "/etc/localtime:/etc/localtime:ro" - "/opt/hyhal:/opt/hyhal" - "/data/deploy/ds70/ds70:/data/ds70" - "/data/deploy/models:/data/models" group_add: - video environment: - LANG=zh_CN.UTF-8 - LANGUAGE=zh_CN.UTF-8 cap_add: - SYS_PTRACE security_opt: - seccomp=unconfined ulimits: memlock: soft: -1 hard: -1 command: /bin/bash tty: true stdin_open: true restart: unless-stopped vllm启动： nohup vllm serve /data/models/Qwen3.5-122B-A10B --port 16663 --tensor-parallel-size 8 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --disable-custom-all-reduce --gpu-memory-utilization 0.925 --default-chat-template-kwargs '{"enable_thinking": true}' --served-model-name Qwen3.5-122B-A10B & 然后通过ragflow接入大模型，在进行问答过程中发现没有返回思考过程返回到前端。 通过以下脚本直接调用，可以看到think过程 from openai import OpenAI client = OpenAI( base_url="http://192.168.0.103:16663/v1", api_key="anything" # vLLM 不需要真实 key ) response = client.chat.completions.create( model="Qwen3.5-122B-A10B", messages=[{"role": "user", "content": "你的问题"}], temperature=1.0, max_tokens=8192 ) print("=== 思考过程 ===") print(response.choices[0].message.reasoning or "无思考过程") print("\n=== 最终答案 ===") print(response.choices[0].message.content)