This commit is contained in:
parent
8a0217e8dc
commit
fdf3422545
30
Dockerfile
30
Dockerfile
|
|
@ -1,18 +1,18 @@
|
||||||
############################################################
|
############################################################
|
||||||
# Stage-0: 构建依赖轮子(PyTorch + SGLang + sgl_kernel) #
|
# Stage-0: 构建依赖轮子(PyTorch + SGLang + sgl_kernel) #
|
||||||
############################################################
|
############################################################
|
||||||
ARG CUDA_VERSION=12.8.1 # 仅 12.6.1 / 12.8.1 测过
|
ARG CUDA_VERSION=12.8.1
|
||||||
FROM nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu22.04 AS builder
|
FROM nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu22.04 AS builder
|
||||||
|
|
||||||
# ---- Python 环境 ----
|
# ---- Python 环境 ----
|
||||||
RUN apt-get update && \
|
RUN apt-get update && \
|
||||||
apt-get install -y --no-install-recommends python3 python3-pip && \
|
apt-get install -y --no-install-recommends python3 python3-pip python3-distutils && \
|
||||||
ln -sf /usr/bin/python3 /usr/bin/python && \
|
ln -sf /usr/bin/python3 /usr/bin/python && \
|
||||||
python -m pip install --no-cache-dir --upgrade pip
|
python -m pip install --no-cache-dir --upgrade pip setuptools wheel html5lib six
|
||||||
|
|
||||||
# ---- PyTorch / torchvision / SGLang / sgl_kernel ----
|
# ---- PyTorch / torchvision / SGLang / sgl_kernel ----
|
||||||
ARG TORCH_VER=2.7.1 # cu128 轮子从 2.7.1 起提供
|
ARG TORCH_VER=2.7.1
|
||||||
ARG TV_VER=0.22.1 # 与 torch-2.7.1 对齐
|
ARG TV_VER=0.22.1
|
||||||
RUN case "$CUDA_VERSION" in \
|
RUN case "$CUDA_VERSION" in \
|
||||||
12.6.1) CUINDEX=126 ;; \
|
12.6.1) CUINDEX=126 ;; \
|
||||||
12.8.1) CUINDEX=128 ;; \
|
12.8.1) CUINDEX=128 ;; \
|
||||||
|
|
@ -22,11 +22,15 @@ RUN case "$CUDA_VERSION" in \
|
||||||
torch==${TORCH_VER}+cu${CUINDEX} \
|
torch==${TORCH_VER}+cu${CUINDEX} \
|
||||||
torchvision==${TV_VER}+cu${CUINDEX} \
|
torchvision==${TV_VER}+cu${CUINDEX} \
|
||||||
--extra-index-url https://download.pytorch.org/whl/cu${CUINDEX} && \
|
--extra-index-url https://download.pytorch.org/whl/cu${CUINDEX} && \
|
||||||
python -m pip install --no-cache-dir sglang==0.4.8.post1 && \
|
python -m pip install --no-cache-dir \
|
||||||
# NCCL 2.27——CUDA 12.x 官方推荐
|
sglang==0.4.8.post1 \
|
||||||
python -m pip install --no-cache-dir nvidia-nccl-cu12==2.27.3 --force-reinstall --no-deps && \
|
sgl-kernel==0.0.2.post17 \
|
||||||
# ✅ 官方预编译,带 cp310 & cu128 标签;不再手写直链
|
nvidia-nccl-cu12==2.27.3 --force-reinstall --no-deps && \
|
||||||
python -m pip install --no-cache-dir sgl-kernel==0.0.2.post17
|
# ✅ 补全依赖(必须)
|
||||||
|
python -m pip install --no-cache-dir \
|
||||||
|
pydantic psutil pyzmq pynvml transformers==4.48.3 uvicorn fastapi && \
|
||||||
|
# ✅ 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 #
|
# Stage-1: 生成最小运行镜像 #
|
||||||
|
|
@ -42,14 +46,14 @@ RUN apt-get update && \
|
||||||
ln -sf /usr/bin/python3 /usr/bin/python && \
|
ln -sf /usr/bin/python3 /usr/bin/python && \
|
||||||
rm -rf /var/lib/apt/lists/*
|
rm -rf /var/lib/apt/lists/*
|
||||||
|
|
||||||
# ---- 拷贝依赖 ----
|
# ---- 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/lib/python3.10/dist-packages /usr/local/lib/python3.10/dist-packages
|
||||||
COPY --from=builder /usr/local/bin /usr/local/bin
|
COPY --from=builder /usr/local/bin /usr/local/bin
|
||||||
|
|
||||||
# ---- 拷贝模型(改成你的实际路径即可) ----
|
# ---- 拷贝模型(路径可换) ----
|
||||||
COPY ./Alibaba/Qwen3-8B /root/.cradle/Alibaba/Qwen3-8B
|
COPY ./Alibaba/Qwen3-8B /root/.cradle/Alibaba/Qwen3-8B
|
||||||
|
|
||||||
# ---- 端口 & 启动 ----
|
# ---- 启动服务 ----
|
||||||
EXPOSE 30000
|
EXPOSE 30000
|
||||||
CMD ["python3", "-m", "sglang.launch_server", \
|
CMD ["python3", "-m", "sglang.launch_server", \
|
||||||
"--host", "0.0.0.0", \
|
"--host", "0.0.0.0", \
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue