.

Dockerfile

@@ -72,15 +72,11 @@ RUN pip install . && \
     python3 -m pip wheel . --no-deps -w dist/
 
 
-# builder-extras 阶段：补充工具链 + 安装 vllm
+# ── 安装 vllm（跳过编译，直接装） ─────────────────────────────────────────────
 WORKDIR /opt
-RUN apt-get update && apt-get install -y --no-install-recommends \
-    ninja-build cmake build-essential python3-dev git && \
-    pip install setuptools wheel setuptools_scm && \
-    git clone -b v0.9.1 https://github.com/vllm-project/vllm.git && \
-    cd vllm && \
-    python3 -m pip install . --no-build-isolation --no-deps && \
-    python3 -m pip wheel . --no-deps -w /tmp/vllm_wheels
+RUN pip install setuptools wheel setuptools_scm && \
+    pip install git+https://github.com/vllm-project/vllm.git@v0.9.1 --no-deps && \
+    python3 -m pip wheel vllm -w /tmp/vllm_wheels --no-deps
 
 
 # ── 编译你本地 sglang 源码并打 wheel ───────────────────────────────────────
@@ -122,6 +118,7 @@ COPY --from=builder-extras /wheels /tmp/wheels
 #RUN python3 -m pip install --no-cache-dir /tmp/wheels/* && rm -rf /tmp/wheels
 # ✅ 优先装你自编的 torch，避免被 PyPI 上的覆盖
 RUN python3 -m pip install --no-cache-dir /tmp/wheels/torch*.whl && \
+    python3 -m pip install /tmp/wheels/vllm-*.whl && \
     python3 -m pip install --no-cache-dir /tmp/wheels/* && \
     python3 -c "from torch.distributed import Backend; print('✅ Runtime torch distributed OK, GLOO =', Backend.GLOO)" && \
     rm -rf /tmp/wheels