56 lines
2.6 KiB
Docker
56 lines
2.6 KiB
Docker
###############################################################################
|
||
# Stage 0 — builder：CUDA 12.6.1 + nvcc/gcc，用来
|
||
# 1) 下载 torch-2.7.1+cu124 / torchvision-0.22.1+cu124 wheel
|
||
# 2) 下载 flashinfer-python (torch2.7/cu124) wheel
|
||
# 3) 编译本地 sglang 源码并打 wheel
|
||
###############################################################################
|
||
ARG CUDA_VERSION=12.6.1
|
||
ARG CUINDEX=124 # 使用 cu124 官方 wheel
|
||
FROM nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04 AS builder
|
||
|
||
ENV DEBIAN_FRONTEND=noninteractive \
|
||
PYTHONUNBUFFERED=1 LANG=C.UTF-8 LC_ALL=C.UTF-8
|
||
|
||
RUN apt-get update && apt-get install -y --no-install-recommends \
|
||
python3 python3-pip python3-distutils build-essential git ca-certificates && \
|
||
python3 -m pip install --no-cache-dir --upgrade pip setuptools wheel
|
||
|
||
# ---- ① 预下载 PyTorch / TorchVision cu124 wheel --------------------------
|
||
RUN python3 -m pip download --no-deps -d /tmp/wheels \
|
||
--index-url https://download.pytorch.org/whl/cu${CUINDEX} \
|
||
torch==2.7.1 torchvision==0.22.1
|
||
|
||
# ---- ② 预下载 flashinfer (torch2.7 / cu124) ------------------------------
|
||
RUN python3 -m pip download --no-deps -d /tmp/wheels \
|
||
--find-links https://flashinfer.ai/whl/cu${CUINDEX}/torch2.7/flashinfer-python \
|
||
flashinfer-python
|
||
|
||
# ---- ③ 安装依赖，供后续 sglang 编译 --------------------------------------
|
||
RUN python3 -m pip install --no-cache-dir /tmp/wheels/*
|
||
|
||
# ---- ④ COPY 本地 sglang 源码并编 wheel ------------------------------------
|
||
COPY ./sglang /sgl-workspace/sglang
|
||
WORKDIR /sgl-workspace/sglang/python
|
||
RUN python3 -m pip wheel '.[srt,openai]' --no-deps -w /tmp/wheels
|
||
|
||
###############################################################################
|
||
# Stage 1 — runtime：极简运行镜像（无编译链，离线装 wheel）
|
||
###############################################################################
|
||
ARG CUDA_VERSION=12.6.1
|
||
FROM nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu22.04
|
||
|
||
ENV DEBIAN_FRONTEND=noninteractive \
|
||
PYTHONUNBUFFERED=1 LANG=C.UTF-8 LC_ALL=C.UTF-8
|
||
|
||
RUN apt-get update && apt-get install -y --no-install-recommends \
|
||
python3 python3-pip python3-distutils ca-certificates && \
|
||
rm -rf /var/lib/apt/lists/* && \
|
||
python3 -m pip install --no-cache-dir --upgrade pip
|
||
|
||
# ---- 一次性离线安装 torch / TV / flashinfer / sglang ----------------------
|
||
COPY --from=builder /tmp/wheels /tmp/wheels
|
||
RUN python3 -m pip install --no-cache-dir /tmp/wheels/* && rm -rf /tmp/wheels
|
||
|
||
# ---- 仅打印帮助（CPU 机器也能跑） ----------------------------------------
|
||
CMD ["python3", "-m", "sglang.launch_server", "--help"]
|