diff --git a/docker/Dockerfile b/docker/Dockerfile index 11eac04..05bcf55 100644 --- a/docker/Dockerfile +++ b/docker/Dockerfile @@ -286,22 +286,20 @@ COPY scripts/cache_warmer.py /opt/scripts/cache_warmer.py # =========================================================================== RUN set -ex && \ export http_proxy=${GIT_PROXY} https_proxy=${GIT_PROXY} && \ - # 先安装 vllm 的全部 Python 运行时依赖(--no-build-isolation 不自动装) - # 跳过 setuptools 版本约束(与 PyTorch ROCm 构建工具链冲突) + # 1) 安装 vllm(让 pip 完整解析依赖,参考 ROCm7.2升级指南.md 7.8 节) cd /opt/vllm && \ - grep -v 'setuptools' requirements/common.txt > /tmp/vllm_deps.txt && \ - ${VENV}/bin/pip install -r /tmp/vllm_deps.txt && \ ${VENV}/bin/pip install -e . --no-build-isolation && \ - ${VENV}/bin/python -c "import torch; v=torch.__version__; assert 'rocm' in v, f'PyTorch overwritten: {v}'; print('PyTorch OK:', v)" && \ + # 2) 装完 vllm 后强制复查 PyTorch(vllm 依赖可能拉成 CUDA 版 torch) ${VENV}/bin/pip uninstall -y triton triton-rocm 2>/dev/null; \ ${VENV}/bin/pip install --force-reinstall \ torch==2.11.0+rocm7.2 torchvision pytorch-triton-rocm \ --index-url ${TORCH_INDEX} && \ - # 深度验证 vllm(确保 AsyncEngineArgs、AsyncLLM 等关键模块可导入,缺包直接构建失败) + ${VENV}/bin/python -c "import torch; v=torch.__version__; assert 'rocm' in v, f'PyTorch overwritten: {v}'; print('PyTorch OK:', v)" && \ + # 3) 深度验证 vllm(确保 AsyncEngineArgs、AsyncLLM 等关键模块可导入,缺包直接构建失败) ${VENV}/bin/python -c "from vllm.engine.arg_utils import AsyncEngineArgs; from vllm.v1.engine.async_llm import AsyncLLM; print('vllm deep import OK:', __import__('vllm').__version__)" && \ - # 确保 mineru-api 子进程不依赖 PYTHONPATH 也能导入 vllm + # 4) 确保 mineru-api 子进程不依赖 PYTHONPATH 也能导入 vllm echo "/opt/vllm" > ${VENV}/lib/python${PYTHON_VER}/site-packages/vllm.pth && \ - # 锁定 ROCm 关键包版本(不含 setuptools,避免与 vllm/mineru 依赖冲突) + # 5) 锁定 ROCm 关键包版本(不含 setuptools,避免与 vllm/mineru 依赖冲突) ${VENV}/bin/pip freeze | grep -iE "^(torch|vllm|triton|pytorch.triton|torchvision|flash.attn|aiter)" >> ${VENV}/torch-constraint.txt && \ rm -rf /opt/vllm_build