diff --git "a/frameworks/vllm/0.16.0/0.16.0GPU\346\265\213\350\257\225.png" "b/frameworks/vllm/0.16.0/0.16.0GPU\346\265\213\350\257\225.png"
new file mode 100644
index 0000000000000000000000000000000000000000..0f765458daa78bf43be16e54577a4fe322cef6ea
Binary files /dev/null and "b/frameworks/vllm/0.16.0/0.16.0GPU\346\265\213\350\257\225.png" differ
diff --git "a/frameworks/vllm/0.16.0/0.16.0\346\227\240GPU\346\265\213\350\257\225.png" "b/frameworks/vllm/0.16.0/0.16.0\346\227\240GPU\346\265\213\350\257\225.png"
new file mode 100644
index 0000000000000000000000000000000000000000..f2b79c32f93e6fa035a9c011a93e7db4317c7a21
Binary files /dev/null and "b/frameworks/vllm/0.16.0/0.16.0\346\227\240GPU\346\265\213\350\257\225.png" differ
diff --git "a/frameworks/vllm/0.16.0/0.16.0\351\225\234\345\203\217\346\236\204\345\273\272.png" "b/frameworks/vllm/0.16.0/0.16.0\351\225\234\345\203\217\346\236\204\345\273\272.png"
new file mode 100644
index 0000000000000000000000000000000000000000..c380adfa473b314bda89b561dc75049522874221
Binary files /dev/null and "b/frameworks/vllm/0.16.0/0.16.0\351\225\234\345\203\217\346\236\204\345\273\272.png" differ
diff --git a/frameworks/vllm/0.16.0/Dockerfile b/frameworks/vllm/0.16.0/Dockerfile
new file mode 100644
index 0000000000000000000000000000000000000000..d18ba8d3967ea164aab9cec18fc161356a617985
--- /dev/null
+++ b/frameworks/vllm/0.16.0/Dockerfile
@@ -0,0 +1,37 @@
+# 基础镜像：OpenCloudOS 9 + CUDA 12.8 开发版，适用于需要 GPU 加速的框架
+FROM opencloudos/opencloudos9-cuda-devel:12.8
+
+# 镜像元信息
+LABEL maintainer="pangxb666"
+LABEL org.opencontainers.image.source="https://gitee.com/OpenCloudOS/ai-agent-container"
+LABEL org.opencontainers.image.description="vLLM 0.16.0 (GPU) on OpenCloudOS 9"
+
+# 安装 Python 3.11 及 pip，安装完成后清理缓存以减小镜像体积
+RUN dnf install -y \
+    python3 \
+    python3-pip \
+    python3-devel \
+    gcc \
+    gcc-c++ \
+    && dnf clean all \
+    && rm -rf /var/cache/yum/*
+
+# 使用清华镜像源安装 vLLM，--extra-index-url 确保拉取 CUDA 12.8 专用 torch wheel
+RUN pip3 install --no-cache-dir vllm==0.16.0 \
+    -i https://pypi.tuna.tsinghua.edu.cn/simple \
+    --trusted-host pypi.tuna.tsinghua.edu.cn \
+    --extra-index-url https://download.pytorch.org/whl/cu128
+
+# 允许容器访问宿主机所有 GPU 设备
+ENV NVIDIA_VISIBLE_DEVICES=all
+# OC9 将 pip 包安装到 lib64，用 ldconfig 批量注册所有 nvidia 动态库路径，避免 libnvshmem_host.so.3 等找不到的问题
+RUN find /usr/local/lib/python3.11/site-packages/nvidia \
+         /usr/local/lib64/python3.11/site-packages/nvidia \
+    -maxdepth 3 -name "*.so*" -exec dirname {} \; 2>/dev/null | sort -u \
+    > /etc/ld.so.conf.d/nvidia-python.conf && ldconfig
+
+# 记录镜像构建时间，便于追溯
+RUN echo $(date +"%Y-%m-%dT%H:%M:%S%z") > /opencloudos_build_date.txt
+
+# 默认进入 Python 3.11 交互环境
+CMD ["python3"]
diff --git a/frameworks/vllm/0.16.0/README.md b/frameworks/vllm/0.16.0/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..2c7a9f54efead1cac03f112e6a2044fa5538608d
--- /dev/null
+++ b/frameworks/vllm/0.16.0/README.md
@@ -0,0 +1,41 @@
+# vLLM 0.16.0 on OpenCloudOS 9
+
+## 基本信息
+
+- **框架版本**：0.16.0
+- **基础镜像**：opencloudos/opencloudos9-cuda-devel:12.8
+- **Python 版本**：3.11
+- **CUDA 版本**：12.8
+- **开源地址**：https://github.com/vllm-project/vllm
+
+## 简介
+
+vLLM 是高吞吐量、低延迟的大语言模型推理与服务引擎，支持 PagedAttention、连续批处理、OpenAI 兼容 API 等特性。
+
+## 构建
+
+```bash
+docker build -t oc9-vllm:0.16.0 .
+```
+
+## 使用示例
+
+```bash
+# 验证版本
+docker run --rm oc9-vllm:0.16.0 \
+    python3.11 -c "import importlib.metadata; print(importlib.metadata.version('vllm'))"
+
+# 运行基础测试（需要 GPU）
+docker run --rm --gpus all oc9-vllm:0.16.0 bash /test.sh
+
+# 启动 OpenAI 兼容推理服务（需要 GPU + 模型）
+docker run --rm --gpus all -p 8000:8000 oc9-vllm:0.16.0 \
+    python3.11 -m vllm.entrypoints.openai.api_server \
+    --model <model_path_or_hf_id> \
+    --port 8000
+```
+
+## 已知问题
+
+- 安装依赖 torch+CUDA wheel，如网络受限需配置代理或使用离线 wheel
+- 完整推理测试需在 GPU 实例上进行，并提前准备好模型文件
diff --git a/frameworks/vllm/0.16.0/build.conf b/frameworks/vllm/0.16.0/build.conf
new file mode 100644
index 0000000000000000000000000000000000000000..cc8fbefad0c8357fc73145cd9d892865242eab59
--- /dev/null
+++ b/frameworks/vllm/0.16.0/build.conf
@@ -0,0 +1,4 @@
+# vllm 0.16.0 on OpenCloudOS 9 (GPU)
+IMAGE_NAME=oc9-vllm
+IMAGE_TAG=0.16.0
+GPU_TEST=true
diff --git a/frameworks/vllm/0.16.0/test.sh b/frameworks/vllm/0.16.0/test.sh
new file mode 100644
index 0000000000000000000000000000000000000000..3e2eeaac9cf6affc6fe6beb72cea02ae54f5f0f5
--- /dev/null
+++ b/frameworks/vllm/0.16.0/test.sh
@@ -0,0 +1,101 @@
+#!/bin/bash
+set -e
+
+IMAGE="${1:?ERROR: 缺少镜像参数。用法: bash test.sh <image:tag>}"
+
+echo "=== vLLM 0.16.0 基础功能测试 ==="
+echo "运行时间：$(date '+%Y-%m-%d %H:%M:%S')"
+echo ""
+
+# ============================================================
+# 阶段一：CPU 安全检查（无需 GPU，镜像构建后即可验证）
+# ============================================================
+echo "--- 阶段一：安装验证 ---"
+
+echo -n "检查 vllm 版本... "
+sudo docker run --rm "$IMAGE" python3 -c "
+import importlib.metadata
+version = importlib.metadata.version('vllm')
+print(version)
+assert version == '0.16.0', f'期望 0.16.0，实际 {version}'
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 LLM / SamplingParams 导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm import LLM, SamplingParams
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 AsyncLLMEngine 导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm import AsyncLLMEngine
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 EngineArgs 导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm.engine.arg_utils import EngineArgs
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 OpenAI 兼容入口导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm.entrypoints.openai.chat_completion.serving import OpenAIServingChat
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 SamplingParams 参数构造... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm import SamplingParams
+params = SamplingParams(temperature=0.8, top_p=0.95, max_tokens=128)
+assert params.temperature == 0.8 and params.max_tokens == 128
+print(f'temperature={params.temperature}, max_tokens={params.max_tokens}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 torch CUDA 构建... "
+sudo docker run --rm "$IMAGE" python3 -c "
+import torch
+assert torch.version.cuda is not None, 'torch 未使用 CUDA 构建'
+print(f'torch={torch.__version__}, cuda={torch.version.cuda}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo ""
+
+# ============================================================
+# 阶段二：GPU 运行时检查（需要 NVIDIA GPU 驱动）
+# ============================================================
+echo "--- 阶段二：GPU 运行时检查 ---"
+
+if ! command -v nvidia-smi &>/dev/null; then
+    echo "⚠ 未检测到 nvidia-smi，跳过 GPU 测试（请在 GPU 实例上运行）"
+    echo ""
+    echo "=== 阶段一全部通过，阶段二已跳过（无 GPU 环境） ==="
+    exit 0
+fi
+
+echo -n "检查 nvidia-smi... "
+sudo docker run --rm --gpus all "$IMAGE" nvidia-smi --query-gpu=name,driver_version,memory.total --format=csv,noheader \
+    && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 torch.cuda.is_available()... "
+sudo docker run --rm --gpus all "$IMAGE" python3 -c "
+import torch
+assert torch.cuda.is_available(), 'torch 无法访问 GPU'
+print(f'GPU 数量: {torch.cuda.device_count()}')
+for i in range(torch.cuda.device_count()):
+    print(f'  GPU {i}: {torch.cuda.get_device_name(i)}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 vllm 平台探测... "
+sudo docker run --rm --gpus all "$IMAGE" python3 -c "
+from vllm.platforms import current_platform
+print(f'平台: {type(current_platform).__name__}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo ""
+echo "--- 阶段三：完整推理测试 ---"
+echo "  请使用项目根目录的 cuda_test.sh："
+echo "  bash cuda_test.sh --vllm-version 0.16.0 --vllm-model <model_path>"
+echo ""
+
+echo "=== 所有测试通过 ==="
diff --git "a/frameworks/vllm/0.17.0/0.17.0-vLLM \346\234\215\345\212\241\345\220\257\345\212\250\346\210\220\345\212\237\346\227\245\345\277\227.png" "b/frameworks/vllm/0.17.0/0.17.0-vLLM \346\234\215\345\212\241\345\220\257\345\212\250\346\210\220\345\212\237\346\227\245\345\277\227.png"
new file mode 100644
index 0000000000000000000000000000000000000000..fcd172cb0206f673157294d9868efa666e487265
Binary files /dev/null and "b/frameworks/vllm/0.17.0/0.17.0-vLLM \346\234\215\345\212\241\345\220\257\345\212\250\346\210\220\345\212\237\346\227\245\345\277\227.png" differ
diff --git "a/frameworks/vllm/0.17.0/0.17.0-\346\250\241\345\236\213API \345\257\271\350\257\235\346\265\213\350\257\225.png" "b/frameworks/vllm/0.17.0/0.17.0-\346\250\241\345\236\213API \345\257\271\350\257\235\346\265\213\350\257\225.png"
new file mode 100644
index 0000000000000000000000000000000000000000..9f9f5567e9b731d9f89fdcd583bc6b7e54aa58a5
Binary files /dev/null and "b/frameworks/vllm/0.17.0/0.17.0-\346\250\241\345\236\213API \345\257\271\350\257\235\346\265\213\350\257\225.png" differ
diff --git "a/frameworks/vllm/0.17.0/0.17.0GPU\346\265\213\350\257\225.png" "b/frameworks/vllm/0.17.0/0.17.0GPU\346\265\213\350\257\225.png"
new file mode 100644
index 0000000000000000000000000000000000000000..d5a308b11be2f2b914457ffb026abae06d83a0a4
Binary files /dev/null and "b/frameworks/vllm/0.17.0/0.17.0GPU\346\265\213\350\257\225.png" differ
diff --git "a/frameworks/vllm/0.17.0/0.17.0\346\227\240GPU\346\265\213\350\257\225.png" "b/frameworks/vllm/0.17.0/0.17.0\346\227\240GPU\346\265\213\350\257\225.png"
new file mode 100644
index 0000000000000000000000000000000000000000..4424b6481b1cd39cf9ca743843e0628f2c28d2f0
Binary files /dev/null and "b/frameworks/vllm/0.17.0/0.17.0\346\227\240GPU\346\265\213\350\257\225.png" differ
diff --git "a/frameworks/vllm/0.17.0/0.17.0\351\225\234\345\203\217\346\236\204\345\273\272.png" "b/frameworks/vllm/0.17.0/0.17.0\351\225\234\345\203\217\346\236\204\345\273\272.png"
new file mode 100644
index 0000000000000000000000000000000000000000..3029d7964bb1ac9b884f0b949c637847c1498eb1
Binary files /dev/null and "b/frameworks/vllm/0.17.0/0.17.0\351\225\234\345\203\217\346\236\204\345\273\272.png" differ
diff --git a/frameworks/vllm/0.17.0/Dockerfile b/frameworks/vllm/0.17.0/Dockerfile
new file mode 100644
index 0000000000000000000000000000000000000000..1d6fd87fff520c511acba77476a212f429790bfe
--- /dev/null
+++ b/frameworks/vllm/0.17.0/Dockerfile
@@ -0,0 +1,32 @@
+# 基础镜像：OpenCloudOS 9 + CUDA 12.8 开发版，适用于需要 GPU 加速的框架
+FROM opencloudos/opencloudos9-cuda-devel:12.8
+
+# 镜像元信息
+LABEL maintainer="pangxb666"
+LABEL org.opencontainers.image.source="https://gitee.com/OpenCloudOS/ai-agent-container"
+LABEL org.opencontainers.image.description="vLLM 0.17.0 (GPU) on OpenCloudOS 9"
+
+# 安装 Python 3.11 及 pip，安装完成后清理缓存以减小镜像体积
+RUN dnf install -y \
+    python3 \
+    python3-pip \
+    python3-devel \
+    gcc \
+    gcc-c++ \
+    && dnf clean all \
+    && rm -rf /var/cache/yum/*
+
+# 使用清华镜像源安装 vLLM，--extra-index-url 确保拉取 CUDA 12.8 专用 torch wheel
+RUN pip3 install --no-cache-dir vllm==0.17.0 \
+    -i https://pypi.tuna.tsinghua.edu.cn/simple \
+    --trusted-host pypi.tuna.tsinghua.edu.cn \
+    --extra-index-url https://download.pytorch.org/whl/cu128
+
+# 允许容器访问宿主机所有 GPU 设备
+ENV NVIDIA_VISIBLE_DEVICES=all
+
+# 记录镜像构建时间，便于追溯
+RUN echo $(date +"%Y-%m-%dT%H:%M:%S%z") > /opencloudos_build_date.txt
+
+# 默认进入 Python 3.11 交互环境
+CMD ["python3"]
diff --git a/frameworks/vllm/0.17.0/README.md b/frameworks/vllm/0.17.0/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..a48d1eb03894610189138dfa0746b797e278b0b8
--- /dev/null
+++ b/frameworks/vllm/0.17.0/README.md
@@ -0,0 +1,41 @@
+# vLLM 0.17.0 on OpenCloudOS 9
+
+## 基本信息
+
+- **框架版本**：0.17.0
+- **基础镜像**：opencloudos/opencloudos9-cuda-devel:12.8
+- **Python 版本**：3.11
+- **CUDA 版本**：12.8
+- **开源地址**：https://github.com/vllm-project/vllm
+
+## 简介
+
+vLLM 是高吞吐量、低延迟的大语言模型推理与服务引擎，支持 PagedAttention、连续批处理、OpenAI 兼容 API 等特性。
+
+## 构建
+
+```bash
+docker build -t oc9-vllm:0.17.0 .
+```
+
+## 使用示例
+
+```bash
+# 验证版本
+docker run --rm oc9-vllm:0.17.0 \
+    python3.11 -c "import importlib.metadata; print(importlib.metadata.version('vllm'))"
+
+# 运行基础测试（需要 GPU）
+docker run --rm --gpus all oc9-vllm:0.17.0 bash /test.sh
+
+# 启动 OpenAI 兼容推理服务（需要 GPU + 模型）
+docker run --rm --gpus all -p 8000:8000 oc9-vllm:0.17.0 \
+    python3.11 -m vllm.entrypoints.openai.api_server \
+    --model <model_path_or_hf_id> \
+    --port 8000
+```
+
+## 已知问题
+
+- 安装依赖 torch+CUDA wheel，如网络受限需配置代理或使用离线 wheel
+- 完整推理测试需在 GPU 实例上进行，并提前准备好模型文件
diff --git a/frameworks/vllm/0.17.0/build.conf b/frameworks/vllm/0.17.0/build.conf
new file mode 100644
index 0000000000000000000000000000000000000000..edd551dda36d82e7ed940707a611398ddd263e0f
--- /dev/null
+++ b/frameworks/vllm/0.17.0/build.conf
@@ -0,0 +1,4 @@
+# vllm 0.17.0 on OpenCloudOS 9 (GPU)
+IMAGE_NAME=oc9-vllm
+IMAGE_TAG=0.17.0
+GPU_TEST=true
diff --git a/frameworks/vllm/0.17.0/test.sh b/frameworks/vllm/0.17.0/test.sh
new file mode 100644
index 0000000000000000000000000000000000000000..0f488aac55a6d5ca68191c631f1e01f32f499a2f
--- /dev/null
+++ b/frameworks/vllm/0.17.0/test.sh
@@ -0,0 +1,101 @@
+#!/bin/bash
+set -e
+
+IMAGE="${1:?ERROR: 缺少镜像参数。用法: bash test.sh <image:tag>}"
+
+echo "=== vLLM 0.17.0 基础功能测试 ==="
+echo "运行时间：$(date '+%Y-%m-%d %H:%M:%S')"
+echo ""
+
+# ============================================================
+# 阶段一：CPU 安全检查（无需 GPU，镜像构建后即可验证）
+# ============================================================
+echo "--- 阶段一：安装验证 ---"
+
+echo -n "检查 vllm 版本... "
+sudo docker run --rm "$IMAGE" python3 -c "
+import importlib.metadata
+version = importlib.metadata.version('vllm')
+print(version)
+assert version == '0.17.0', f'期望 0.17.0，实际 {version}'
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 LLM / SamplingParams 导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm import LLM, SamplingParams
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 AsyncLLMEngine 导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm import AsyncLLMEngine
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 EngineArgs 导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm.engine.arg_utils import EngineArgs
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 OpenAI 兼容入口导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm.entrypoints.openai.chat_completion.serving import OpenAIServingChat
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 SamplingParams 参数构造... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm import SamplingParams
+params = SamplingParams(temperature=0.8, top_p=0.95, max_tokens=128)
+assert params.temperature == 0.8 and params.max_tokens == 128
+print(f'temperature={params.temperature}, max_tokens={params.max_tokens}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 torch CUDA 构建... "
+sudo docker run --rm "$IMAGE" python3 -c "
+import torch
+assert torch.version.cuda is not None, 'torch 未使用 CUDA 构建'
+print(f'torch={torch.__version__}, cuda={torch.version.cuda}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo ""
+
+# ============================================================
+# 阶段二：GPU 运行时检查（需要 NVIDIA GPU 驱动）
+# ============================================================
+echo "--- 阶段二：GPU 运行时检查 ---"
+
+if ! command -v nvidia-smi &>/dev/null; then
+    echo "⚠ 未检测到 nvidia-smi，跳过 GPU 测试（请在 GPU 实例上运行）"
+    echo ""
+    echo "=== 阶段一全部通过，阶段二已跳过（无 GPU 环境） ==="
+    exit 0
+fi
+
+echo -n "检查 nvidia-smi... "
+sudo docker run --rm --gpus all "$IMAGE" nvidia-smi --query-gpu=name,driver_version,memory.total --format=csv,noheader \
+    && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 torch.cuda.is_available()... "
+sudo docker run --rm --gpus all "$IMAGE" python3 -c "
+import torch
+assert torch.cuda.is_available(), 'torch 无法访问 GPU'
+print(f'GPU 数量: {torch.cuda.device_count()}')
+for i in range(torch.cuda.device_count()):
+    print(f'  GPU {i}: {torch.cuda.get_device_name(i)}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 vllm 平台探测... "
+sudo docker run --rm --gpus all "$IMAGE" python3 -c "
+from vllm.platforms import current_platform
+print(f'平台: {type(current_platform).__name__}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo ""
+echo "--- 阶段三：完整推理测试 ---"
+echo "  请使用项目根目录的 cuda_test.sh："
+echo "  bash cuda_test.sh --vllm-version 0.17.0 --vllm-model <model_path>"
+echo ""
+
+echo "=== 所有测试通过 ==="
diff --git "a/frameworks/vllm/0.17.1/0.17.1GPU\346\265\213\350\257\225.png" "b/frameworks/vllm/0.17.1/0.17.1GPU\346\265\213\350\257\225.png"
new file mode 100644
index 0000000000000000000000000000000000000000..acb2b66eea236b08b0eaea96615420470d449986
Binary files /dev/null and "b/frameworks/vllm/0.17.1/0.17.1GPU\346\265\213\350\257\225.png" differ
diff --git "a/frameworks/vllm/0.17.1/0.17.1\346\227\240GPU\346\265\213\350\257\225.png" "b/frameworks/vllm/0.17.1/0.17.1\346\227\240GPU\346\265\213\350\257\225.png"
new file mode 100644
index 0000000000000000000000000000000000000000..610089807c442ef8e94f0d73513ee07fe48b24ba
Binary files /dev/null and "b/frameworks/vllm/0.17.1/0.17.1\346\227\240GPU\346\265\213\350\257\225.png" differ
diff --git "a/frameworks/vllm/0.17.1/0.17.1\351\225\234\345\203\217\346\236\204\345\273\272.png" "b/frameworks/vllm/0.17.1/0.17.1\351\225\234\345\203\217\346\236\204\345\273\272.png"
new file mode 100644
index 0000000000000000000000000000000000000000..d2723cac295474c75eae42648c8674fd231ca01e
Binary files /dev/null and "b/frameworks/vllm/0.17.1/0.17.1\351\225\234\345\203\217\346\236\204\345\273\272.png" differ
diff --git a/frameworks/vllm/0.17.1/Dockerfile b/frameworks/vllm/0.17.1/Dockerfile
new file mode 100644
index 0000000000000000000000000000000000000000..0898ab69536ffa7a9cf7e87fd5f630817d6436d8
--- /dev/null
+++ b/frameworks/vllm/0.17.1/Dockerfile
@@ -0,0 +1,32 @@
+# 基础镜像：OpenCloudOS 9 + CUDA 12.8 开发版，适用于需要 GPU 加速的框架
+FROM opencloudos/opencloudos9-cuda-devel:12.8
+
+# 镜像元信息
+LABEL maintainer="pangxb666"
+LABEL org.opencontainers.image.source="https://gitee.com/OpenCloudOS/ai-agent-container"
+LABEL org.opencontainers.image.description="vLLM 0.17.1 (GPU) on OpenCloudOS 9"
+
+# 安装 Python 3.11 及 pip，安装完成后清理缓存以减小镜像体积
+RUN dnf install -y \
+    python3 \
+    python3-pip \
+    python3-devel \
+    gcc \
+    gcc-c++ \
+    && dnf clean all \
+    && rm -rf /var/cache/yum/*
+
+# 使用清华镜像源安装 vLLM，--extra-index-url 确保拉取 CUDA 12.8 专用 torch wheel
+RUN pip3 install --no-cache-dir vllm==0.17.1 \
+    -i https://pypi.tuna.tsinghua.edu.cn/simple \
+    --trusted-host pypi.tuna.tsinghua.edu.cn \
+    --extra-index-url https://download.pytorch.org/whl/cu128
+
+# 允许容器访问宿主机所有 GPU 设备
+ENV NVIDIA_VISIBLE_DEVICES=all
+
+# 记录镜像构建时间，便于追溯
+RUN echo $(date +"%Y-%m-%dT%H:%M:%S%z") > /opencloudos_build_date.txt
+
+# 默认进入 Python 3.11 交互环境
+CMD ["python3"]
diff --git a/frameworks/vllm/0.17.1/README.md b/frameworks/vllm/0.17.1/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..41574244ccb7bec0bce73c92a65530a4b06235a3
--- /dev/null
+++ b/frameworks/vllm/0.17.1/README.md
@@ -0,0 +1,41 @@
+# vLLM 0.17.1 on OpenCloudOS 9
+
+## 基本信息
+
+- **框架版本**：0.17.1
+- **基础镜像**：opencloudos/opencloudos9-cuda-devel:12.8
+- **Python 版本**：3.11
+- **CUDA 版本**：12.8
+- **开源地址**：https://github.com/vllm-project/vllm
+
+## 简介
+
+vLLM 是高吞吐量、低延迟的大语言模型推理与服务引擎，支持 PagedAttention、连续批处理、OpenAI 兼容 API 等特性。
+
+## 构建
+
+```bash
+docker build -t oc9-vllm:0.17.1 .
+```
+
+## 使用示例
+
+```bash
+# 验证版本
+docker run --rm oc9-vllm:0.17.1 \
+    python3.11 -c "import importlib.metadata; print(importlib.metadata.version('vllm'))"
+
+# 运行基础测试（需要 GPU）
+docker run --rm --gpus all oc9-vllm:0.17.1 bash /test.sh
+
+# 启动 OpenAI 兼容推理服务（需要 GPU + 模型）
+docker run --rm --gpus all -p 8000:8000 oc9-vllm:0.17.1 \
+    python3.11 -m vllm.entrypoints.openai.api_server \
+    --model <model_path_or_hf_id> \
+    --port 8000
+```
+
+## 已知问题
+
+- 安装依赖 torch+CUDA wheel，如网络受限需配置代理或使用离线 wheel
+- 完整推理测试需在 GPU 实例上进行，并提前准备好模型文件
diff --git a/frameworks/vllm/0.17.1/build.conf b/frameworks/vllm/0.17.1/build.conf
new file mode 100644
index 0000000000000000000000000000000000000000..3d14035ea2b73b8ee0687cf599b3d1a355094cf2
--- /dev/null
+++ b/frameworks/vllm/0.17.1/build.conf
@@ -0,0 +1,4 @@
+# vllm 0.17.1 on OpenCloudOS 9 (GPU)
+IMAGE_NAME=oc9-vllm
+IMAGE_TAG=0.17.1
+GPU_TEST=true
diff --git a/frameworks/vllm/0.17.1/test.sh b/frameworks/vllm/0.17.1/test.sh
new file mode 100644
index 0000000000000000000000000000000000000000..23abad8136b08e0a8d1a3eec5a445c939d3a926a
--- /dev/null
+++ b/frameworks/vllm/0.17.1/test.sh
@@ -0,0 +1,101 @@
+#!/bin/bash
+set -e
+
+IMAGE="${1:?ERROR: 缺少镜像参数。用法: bash test.sh <image:tag>}"
+
+echo "=== vLLM 0.17.1 基础功能测试 ==="
+echo "运行时间：$(date '+%Y-%m-%d %H:%M:%S')"
+echo ""
+
+# ============================================================
+# 阶段一：CPU 安全检查（无需 GPU，镜像构建后即可验证）
+# ============================================================
+echo "--- 阶段一：安装验证 ---"
+
+echo -n "检查 vllm 版本... "
+sudo docker run --rm "$IMAGE" python3 -c "
+import importlib.metadata
+version = importlib.metadata.version('vllm')
+print(version)
+assert version == '0.17.1', f'期望 0.17.1，实际 {version}'
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 LLM / SamplingParams 导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm import LLM, SamplingParams
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 AsyncLLMEngine 导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm import AsyncLLMEngine
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 EngineArgs 导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm.engine.arg_utils import EngineArgs
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 OpenAI 兼容入口导入... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm.entrypoints.openai.chat_completion.serving import OpenAIServingChat
+print('ok')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 SamplingParams 参数构造... "
+sudo docker run --rm "$IMAGE" python3 -c "
+from vllm import SamplingParams
+params = SamplingParams(temperature=0.8, top_p=0.95, max_tokens=128)
+assert params.temperature == 0.8 and params.max_tokens == 128
+print(f'temperature={params.temperature}, max_tokens={params.max_tokens}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 torch CUDA 构建... "
+sudo docker run --rm "$IMAGE" python3 -c "
+import torch
+assert torch.version.cuda is not None, 'torch 未使用 CUDA 构建'
+print(f'torch={torch.__version__}, cuda={torch.version.cuda}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo ""
+
+# ============================================================
+# 阶段二：GPU 运行时检查（需要 NVIDIA GPU 驱动）
+# ============================================================
+echo "--- 阶段二：GPU 运行时检查 ---"
+
+if ! command -v nvidia-smi &>/dev/null; then
+    echo "⚠ 未检测到 nvidia-smi，跳过 GPU 测试（请在 GPU 实例上运行）"
+    echo ""
+    echo "=== 阶段一全部通过，阶段二已跳过（无 GPU 环境） ==="
+    exit 0
+fi
+
+echo -n "检查 nvidia-smi... "
+sudo docker run --rm --gpus all "$IMAGE" nvidia-smi --query-gpu=name,driver_version,memory.total --format=csv,noheader \
+    && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 torch.cuda.is_available()... "
+sudo docker run --rm --gpus all "$IMAGE" python3 -c "
+import torch
+assert torch.cuda.is_available(), 'torch 无法访问 GPU'
+print(f'GPU 数量: {torch.cuda.device_count()}')
+for i in range(torch.cuda.device_count()):
+    print(f'  GPU {i}: {torch.cuda.get_device_name(i)}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo -n "检查 vllm 平台探测... "
+sudo docker run --rm --gpus all "$IMAGE" python3 -c "
+from vllm.platforms import current_platform
+print(f'平台: {type(current_platform).__name__}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo ""
+echo "--- 阶段三：完整推理测试 ---"
+echo "  请使用项目根目录的 cuda_test.sh："
+echo "  bash cuda_test.sh --vllm-version 0.17.1 --vllm-model <model_path>"
+echo ""
+
+echo "=== 所有测试通过 ==="