目录千问3vllm部署方法：qwen2.5:14bOpenGVLabinternvl3_5-8b 推荐vllm0.14,自动安装了torch2.9 推理ok腾讯tencent/HY-MT1.5-7B-FP8 推荐HY-MT1.5-1.8B和HY-MT1.5-7B-GPTQ-Int4比较HY-MT1.5-7B-GPTQ-Int4 400ms网易 netease/youdao_Confucius3-MathL20服务器，推理时间20s，ZAYA1-8Bhttps://huggingface.co/Zyphra/ZAYA1-8Bhf download Zyphra/ZAYA1-8B --local-dir ./Zyphra_ZAYA1-8B千问3Qwen/Qwen3-235B-A22B-GPTQ-Int4swift/Qwen3-235B-A22B-Thinking-2507-AWQvllm部署方法：vllm serve /data/lbg/models/Qwen_Qwen3-235B-A22B-GPTQ-Int4 \ --tensor-parallel-size 8 \ --enable-expert-parallel \ --max-model-len 4096 \ --gpu-memory-utilization 0.6 \ --disable-cuda-grap