.

Dockerfile

@@ -1,7 +1,7 @@
 ############################
-#  Stage-0: 构建镜像        #
+#  Stage-0: 构建依赖轮子    #
 ############################
-ARG CUDA_VERSION=12.8.1          # 仅 12.6.1 / 12.8.1 经测试可用
+ARG CUDA_VERSION=12.8.1
 FROM nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu22.04 AS builder
 
 # ---- Python & pip ----
@@ -10,9 +10,9 @@ RUN apt-get update && \
     ln -sf /usr/bin/python3 /usr/bin/python && \
     python -m pip install --no-cache-dir --upgrade pip
 
-# ---- 安装 PyTorch + SGLang ----
-ARG TORCH_VER=2.7.1              # cu128 轮子从 2.7.* 开始提供
-ARG TV_VER=0.22.1                # 0.22.1 与 torch-2.7.1 完全兼容
+# ---- PyTorch + SGLang ----
+ARG TORCH_VER=2.7.1
+ARG TV_VER=0.22.1
 RUN case "$CUDA_VERSION" in \
         12.6.1) CUINDEX=126 ;; \
         12.8.1) CUINDEX=128 ;; \
@@ -23,11 +23,10 @@ RUN case "$CUDA_VERSION" in \
         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 && \
-    # 12.8 额外装官方 sgl_kernel & NCCL
+    # CUDA 12.x 需要新版 NCCL
     python -m pip install --no-cache-dir nvidia-nccl-cu12==2.27.3 --force-reinstall --no-deps && \
-    python -m pip install --no-cache-dir \
-        https://github.com/sgl-project/whl/releases/download/v0.2.0/sgl_kernel-0.2.0+cu128-cp310-abi3-manylinux2014_x86_64.whl \
-        --force-reinstall --no-deps
+    # ✅ 直接用 PyPI 上带 cp310 标签的二进制轮子，省去编译
+    python -m pip install --no-cache-dir sgl-kernel==0.0.2.post17
 
 ############################
 #  Stage-1: 运行镜像        #
@@ -37,23 +36,20 @@ FROM nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu22.04
 ENV DEBIAN_FRONTEND=noninteractive \
     PYTHONUNBUFFERED=1
 
-# ---- 基础 Python ----
 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/*
 
-# ---- 复制依赖轮子 ----
+# 复制依赖
 COPY --from=builder /usr/local/lib/python3.*/dist-packages /usr/local/lib/python3.*/dist-packages
 COPY --from=builder /usr/local/bin /usr/local/bin
 
-# ---- 拷贝模型权重 ----
+# 拷贝模型（示例）
 COPY ./Alibaba/Qwen3-8B /root/.cradle/Alibaba/Qwen3-8B
 
-# ---- 暴露端口 ----
 EXPOSE 30000
 
-# ---- 启动命令 ----
 CMD ["python3", "-m", "sglang.launch_server", \
      "--host", "0.0.0.0", \
      "--port", "30000", \