diff --git a/frameworks/tensorrt-llm/1.2.0/Dockerfile b/frameworks/tensorrt-llm/1.2.0/Dockerfile
new file mode 100644
index 0000000000000000000000000000000000000000..2e5b77de0ca82098975888b8c94ed39fde3c310a
--- /dev/null
+++ b/frameworks/tensorrt-llm/1.2.0/Dockerfile
@@ -0,0 +1,59 @@
+# syntax=docker/dockerfile:1
+
+FROM opencloudos/opencloudos9-cuda-devel:12.8
+
+LABEL maintainer="OpenCloudOS AI Community"
+LABEL org.opencontainers.image.source="https://gitee.com/OpenCloudOS/ai-agent-container"
+LABEL org.opencontainers.image.description="TensorRT-LLM 1.2.0 (GPU) on OpenCloudOS 9"
+
+# Install system dependencies
+RUN --mount=type=cache,target=/var/cache/dnf \
+    dnf install -y \
+    python3.12 \
+    python3.12-devel \
+    openmpi \
+    openmpi-devel \
+    && dnf clean all
+
+# Set environment variables
+ENV NVIDIA_VISIBLE_DEVICES=all
+ENV PATH=/usr/lib64/openmpi/bin:$PATH
+ENV LD_LIBRARY_PATH=/usr/lib64:/usr/lib64/cuda-compat-12-8:/usr/lib64/openmpi/lib:/usr/lib/python3.12/site-packages/nvidia/nvshmem/lib:/usr/lib64/cuda:/usr/lib64/cuda/lib64:/usr/local/cuda/lib64
+
+# Configure pip: parallel downloads + prefer pre-built wheels
+ENV PIP_CONCURRENT_DOWNLOADS=5
+
+# Install pip for python3.12, then install tensorrt-llm from NVIDIA PyPI
+RUN python3.12 -m ensurepip --upgrade \
+    && python3.12 -m pip install --no-cache-dir --upgrade pip setuptools wheel
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    for i in 1 2 3; do \
+        python3.12 -m pip install --prefer-binary \
+            nvidia-cublas \
+            torch \
+            tensorrt_llm==1.2.0 \
+            --extra-index-url https://pypi.nvidia.com/ \
+        && break || (echo "pip install attempt $i failed, retrying in 10s..." && sleep 10); \
+    done; \
+    python3.12 -m pip show tensorrt_llm > /dev/null || { echo "ERROR: tensorrt_llm not installed after 3 attempts"; exit 1; }
+
+# Register nvidia pip package libraries (CUDA 13 runtime libs from pip)
+RUN touch /etc/ld.so.conf.d/nvidia-pip.conf \
+    && for dir in $(find / -path "*/nvidia/*/lib" -type d 2>/dev/null); do \
+        echo "$dir" >> /etc/ld.so.conf.d/nvidia-pip.conf; \
+    done \
+    && cat /etc/ld.so.conf.d/nvidia-pip.conf \
+    && ldconfig
+
+# Fix: OpenCloudOS Python reports version "3.12.12+" which breaks packaging.version.parse
+RUN DATASETS_CFG="$(python3.12 -c 'import importlib.util as u; s=u.find_spec("datasets"); print(s.submodule_search_locations[0]+"/config.py") if s else ""')" \
+    && if [ -n "$DATASETS_CFG" ]; then \
+        sed -i "s/platform.python_version()/platform.python_version().split('+')[0]/" "$DATASETS_CFG"; \
+    else \
+        echo "datasets not installed, skipping version fix"; \
+    fi
+
+RUN echo $(date +"%Y-%m-%dT%H:%M:%S%z") > /opencloudos_build_date.txt
+
+CMD ["python3.12"]
diff --git a/frameworks/tensorrt-llm/1.2.0/README.md b/frameworks/tensorrt-llm/1.2.0/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..cd13a639bc3d3095e72c4cdbbe21474360143711
--- /dev/null
+++ b/frameworks/tensorrt-llm/1.2.0/README.md
@@ -0,0 +1,54 @@
+# TensorRT-LLM 1.2.0 on OpenCloudOS 9
+
+## 基本信息
+- **框架版本**：1.2.0
+- **基础镜像**：opencloudos/opencloudos9-cuda-devel:12.8
+- **Python 版本**：3.12
+- **CUDA 版本**：12.8（运行时依赖 CUDA 13.x 库由 pip 自动安装）
+
+## 构建
+
+```bash
+docker build -t oc9-tensorrtllm:1.2.0 .
+```
+
+## 使用示例
+
+### 验证安装
+
+```bash
+docker run --rm --gpus all oc9-tensorrtllm:1.2.0 \
+  python3 -c "import tensorrt_llm; print(tensorrt_llm.__version__)"
+```
+
+### 启动推理服务
+
+```bash
+# 1. 下载模型权重
+sudo mkdir -p /models
+sudo git clone https://www.modelscope.cn/Qwen/Qwen2.5-7B-Instruct.git /models/Qwen2.5-7B-Instruct
+
+# 2. 启动容器
+docker run -itd --name tensorrt_serving \
+  --gpus all \
+  -p 8000:8000 \
+  -v /models/Qwen2.5-7B-Instruct:/models/Qwen2.5-7B-Instruct \
+  oc9-tensorrtllm:1.2.0 tail -f /dev/null
+
+# 3. 容器内启动推理服务
+docker exec -it tensorrt_serving bash -c \
+  "trtllm-serve /models/Qwen2.5-7B-Instruct --host 0.0.0.0 --port 8000"
+
+# 4. 访问服务
+curl http://localhost:8000/v1/chat/completions \
+  -H 'Content-Type: application/json' \
+  -d '{
+    "model": "Qwen2.5-7B-Instruct",
+    "messages": [{"role":"user","content":"你好"}],
+    "max_tokens": 128
+  }'
+```
+
+## 已知问题
+- 构建时间较长（需从源码编译 TensorRT-LLM）
+- 需要 NVIDIA Driver >= 535，建议使用较新的 GPU 驱动以兼容 CUDA 13.x 运行时
diff --git a/frameworks/tensorrt-llm/1.2.0/build.conf b/frameworks/tensorrt-llm/1.2.0/build.conf
new file mode 100644
index 0000000000000000000000000000000000000000..ec5c4e30984b9557342c883ac7d47b1a5dbbcc01
--- /dev/null
+++ b/frameworks/tensorrt-llm/1.2.0/build.conf
@@ -0,0 +1,4 @@
+# TensorRT-LLM 1.2.0 on OpenCloudOS 9 (GPU)
+IMAGE_NAME=oc9-tensorrtllm
+IMAGE_TAG=1.2.0
+GPU_TEST=true
diff --git a/frameworks/tensorrt-llm/1.2.0/test.sh b/frameworks/tensorrt-llm/1.2.0/test.sh
new file mode 100755
index 0000000000000000000000000000000000000000..a4ee1389f93ea79cbd547c90638ac9bdd36335c4
--- /dev/null
+++ b/frameworks/tensorrt-llm/1.2.0/test.sh
@@ -0,0 +1,52 @@
+#!/bin/bash
+set -e
+
+IMAGE="${1:?ERROR: 缺少镜像参数。用法: bash test.sh <image:tag>}"
+
+echo "=== TensorRT-LLM 1.2.0 镜像功能测试 ==="
+
+# 1. Verify CUDA environment
+echo -n "检查 CUDA 环境... "
+docker run --rm --gpus all "$IMAGE" python3.12 -c "
+import torch
+assert torch.cuda.is_available(), 'CUDA not available'
+print(f'GPU: {torch.cuda.get_device_name(0)}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+# 2. Verify tensorrt_llm import
+echo -n "检查 tensorrt_llm import... "
+docker run --rm --gpus all "$IMAGE" python3.12 -c "
+import tensorrt_llm
+print(f'tensorrt_llm version: {tensorrt_llm.__version__}')
+assert tensorrt_llm.__version__ == '1.2.0', f'version mismatch: {tensorrt_llm.__version__}'
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+# 3. Verify trtllm-serve command
+echo -n "检查 trtllm-serve 命令... "
+docker run --rm --gpus all "$IMAGE" python3.12 -c "
+import subprocess, sys
+result = subprocess.run(['trtllm-serve', '--help'], capture_output=True, text=True)
+if result.returncode == 0:
+    print('trtllm-serve 命令可用')
+else:
+    print(f'trtllm-serve 异常: {result.stderr[:200]}')
+    sys.exit(1)
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+# 4. Verify TensorRT version
+echo -n "检查 TensorRT 版本... "
+docker run --rm --gpus all "$IMAGE" python3.12 -c "
+import tensorrt as trt
+print(f'TensorRT version: {trt.__version__}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+# 5. Verify key dependencies
+echo -n "检查关键依赖... "
+docker run --rm --gpus all "$IMAGE" python3.12 -c "
+import torch, transformers, numpy as np
+print(f'PyTorch: {torch.__version__}')
+print(f'Transformers: {transformers.__version__}')
+print(f'NumPy: {np.__version__}')
+" && echo "✓ 通过" || { echo "✗ 失败"; exit 1; }
+
+echo "=== 所有测试通过 ==="
diff --git a/frameworks/tensorrt-llm/1.2.0/test_result.png b/frameworks/tensorrt-llm/1.2.0/test_result.png
new file mode 100644
index 0000000000000000000000000000000000000000..f8e5e6bde61b6566ae639765fc8c913ff3aeeb4a
Binary files /dev/null and b/frameworks/tensorrt-llm/1.2.0/test_result.png differ