.

Dockerfile

@@ -89,6 +89,9 @@ WORKDIR /sgl/sglang/python
 RUN python3 -m pip install ".[srt,openai]" --no-build-isolation && \
     python3 -m pip wheel ".[srt,openai]" --no-deps -w /tmp/sg_wheels
 
+# ── 构建 sgl-kernel 的 Python 模块 ───────────────────────────────
+WORKDIR /sgl/sglang/sgl-kernel
+RUN python3 setup.py bdist_wheel && cp dist/sgl_kernel-*.whl /wheels/    
 
 # ── 收集所有 wheel 到 /wheels ──────────────────────────────────────────────
 RUN mkdir -p /wheels && \
@@ -134,6 +137,7 @@ RUN ls -lh /tmp/wheels && \
     rm -f /tmp/wheels/torch-2.7.1a0+*.whl && \
     python3 -m pip install --no-cache-dir --no-deps /tmp/wheels/torch*.whl && \
     python3 -m pip install --no-cache-dir --no-deps /tmp/wheels/vllm-*.whl && \
+    python3 -m pip install --no-cache-dir --no-deps /tmp/wheels/sgl_kernel-*.whl && \
     python3 -m pip install --no-cache-dir --no-deps /tmp/wheels/* && \
     python3 -c "from torch.distributed import Backend; print('✅ Runtime torch distributed OK, GLOO =', Backend.GLOO)" && \
     rm -rf /tmp/wheels
@@ -152,16 +156,16 @@ ADD https://github.com/krallin/tini/releases/download/${TINI_VERSION}/tini /tini
 RUN chmod +x /tini
 ENTRYPOINT ["/tini", "--"]
 
-# ---- 拷贝模型（路径可换） ----
-COPY ./Alibaba/Qwen3-8B /root/.cradle/Alibaba/Qwen3-8B
+# # ---- 拷贝模型（路径可换） ----
+# COPY ./Alibaba/Qwen3-8B /root/.cradle/Alibaba/Qwen3-8B
 
-# ---- 暴露端口 ----
-EXPOSE 30000
+# # ---- 暴露端口 ----
+# EXPOSE 30000
 
-# ---- 启动 SGLang 推理服务 ----
-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"]
+# # ---- 启动 SGLang 推理服务 ----
+# 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"]