64 lines
2.5 KiB
Docker
64 lines
2.5 KiB
Docker
############################################################
|
||
# Stage-0: 构建依赖轮子(PyTorch + SGLang + sgl_kernel) #
|
||
############################################################
|
||
ARG CUDA_VERSION=12.8.1
|
||
FROM nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu22.04 AS builder
|
||
|
||
# ---- Python 环境 ----
|
||
RUN apt-get update && \
|
||
apt-get install -y --no-install-recommends python3 python3-pip python3-distutils && \
|
||
ln -sf /usr/bin/python3 /usr/bin/python && \
|
||
python -m pip install --no-cache-dir --upgrade pip setuptools wheel html5lib six
|
||
|
||
# ---- PyTorch / torchvision / SGLang / sgl_kernel ----
|
||
ARG TORCH_VER=2.7.1
|
||
ARG TV_VER=0.22.1
|
||
RUN case "$CUDA_VERSION" in \
|
||
12.6.1) CUINDEX=126 ;; \
|
||
12.8.1) CUINDEX=128 ;; \
|
||
*) echo "Unsupported CUDA version: $CUDA_VERSION" && exit 1 ;; \
|
||
esac && \
|
||
python -m pip install --no-cache-dir \
|
||
torch==${TORCH_VER}+cu${CUINDEX} \
|
||
torchvision==${TV_VER}+cu${CUINDEX} \
|
||
--extra-index-url https://download.pytorch.org/whl/cu${CUINDEX} && \
|
||
python -m pip install --no-cache-dir \
|
||
sglang==0.4.8.post1 \
|
||
sgl-kernel==0.0.2.post17 \
|
||
nvidia-nccl-cu12==2.27.3 --force-reinstall --no-deps && \
|
||
# ✅ 补全依赖(必须)
|
||
python -m pip install --no-cache-dir \
|
||
pydantic psutil pyzmq pynvml transformers==4.48.3 uvicorn fastapi IPython aiohttp setproctitle orjson uvloop sentencepiece
|
||
# ✅ 测试模块完整性
|
||
#python -c "import sglang, torch, pydantic, transformers, sgl_kernel"
|
||
|
||
############################################################
|
||
# Stage-1: 生成最小运行镜像 #
|
||
############################################################
|
||
FROM nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu22.04
|
||
|
||
ENV DEBIAN_FRONTEND=noninteractive \
|
||
PYTHONUNBUFFERED=1
|
||
|
||
# ---- Python runtime ----
|
||
RUN apt-get update && \
|
||
apt-get install -y --no-install-recommends python3 python3-distutils && \
|
||
ln -sf /usr/bin/python3 /usr/bin/python && \
|
||
rm -rf /var/lib/apt/lists/*
|
||
|
||
# ---- 拷贝 Python 包和入口 ----
|
||
COPY --from=builder /usr/local/lib/python3.10/dist-packages /usr/local/lib/python3.10/dist-packages
|
||
COPY --from=builder /usr/local/bin /usr/local/bin
|
||
|
||
# ---- 拷贝模型(路径可换) ----
|
||
COPY ./Alibaba/Qwen3-8B /root/.cradle/Alibaba/Qwen3-8B
|
||
|
||
# ---- 启动服务 ----
|
||
EXPOSE 30000
|
||
CMD ["python3", "-m", "sglang.launch_server", \
|
||
"--host", "0.0.0.0", \
|
||
"--port", "30000", \
|
||
"--model-path", "/root/.cradle/Alibaba/Qwen3-8B/", \
|
||
"--tp", "1", \
|
||
"--api-key", "token-abc123"]
|