.

Dockerfile

@@ -181,7 +181,8 @@ RUN ldconfig -p | grep -i cupti || (echo "no cupti"; exit 1)
 RUN ldconfig
 
 
-COPY --from=builder-extras /wheels /tmp/wheels
+#COPY --from=builder-extras /wheels /tmp/wheels
+COPY _wheelhouse/ /tmp/wheels/
 
 # ✅ 优先装你自编的 torch，避免被 PyPI 上的覆盖
 RUN ls -lh /tmp/wheels && \
@@ -200,7 +201,7 @@ RUN ls -lh /tmp/wheels && \
 
 # ✅ 安装 Prometheus client
 RUN python3 -m pip install --no-cache-dir prometheus_client
-
+RUN python3 -m pip install --no-deps xgrammar==0.1.24
 # ✅ 设置多进程 metrics 收集目录（用于 MultiProcessCollector）
 ENV PROMETHEUS_MULTIPROC_DIR=/tmp/prometheus