Update sgl-deep-ep release workflow for DeepEP v2 (#34914)

.github/workflows/release-whl-deepep.yml

@@ -42,7 +42,7 @@ jobs:
         include:
           - arch: x86_64
             runner: x64-kernel-build-node
-            source-branch: sgl-deepep-x86
+            source-branch: sgl-deepep-cu12-x86
           - arch: aarch64
             runner: arm-kernel-build-node
             source-branch: sgl-deepep-cu12-arm
@@ -107,10 +107,10 @@ jobs:
         include:
           - arch: x86_64
             runner: x64-kernel-build-node
-            source-branch: sgl-deepep-x86
+            source-branch: sgl-deepep
           - arch: aarch64
             runner: arm-kernel-build-node
-            source-branch: sgl-deepep-arm
+            source-branch: sgl-deepep
     runs-on: ${{ matrix.runner }}
     steps:
       - name: Clean workspace

docker/sgl-deep-ep.Dockerfile

@@ -7,18 +7,16 @@ ARG ARCHITECTURE=x86_64
 ARG CUDA_TAG=cu130
 ARG CUDA_VERSION=13.0
 ARG GDRCOPY_VERSION=2.5.1
+ARG NCCL_VERSION=2.30.7
 ARG PYTHON_TAG=cp312-cp312
 ARG TORCH_VERSION=2.13.0
 
 ENV CUDA_HOME=/usr/local/cuda
-ENV GDRCOPY_HOME=/usr/local
 ENV LD_LIBRARY_PATH=/usr/local/lib:/usr/local/lib64:/usr/local/cuda/lib64:${LD_LIBRARY_PATH}
 ENV PATH=/opt/python/${PYTHON_TAG}/bin:${PATH}
 ENV PYTHON_BIN=/opt/python/${PYTHON_TAG}/bin/python
 
 # These mirror the build and RDMA dependencies used by ci_install_deepep.sh.
-# The image builds only GDRCopy's user-space library: the host owns gdrdrv and
-# passes /dev/gdrdrv through at runtime.
 RUN yum install -y --nogpgcheck --enablerepo=powertools \
         cmake \
         curl \
@@ -51,13 +49,18 @@ RUN set -eux; \
     ln -sf "${cuda_stub}" /usr/lib64/libcuda.so; \
     ln -sf "${cuda_stub}" "/usr/lib/${ARCHITECTURE}-linux-gnu/libcuda.so"
 
-RUN git clone --depth 1 --branch "v${GDRCOPY_VERSION}" \
+# DeepEP v2 uses NCCL Gin on CUDA 13. Keep GDRCopy only for the CUDA 12
-        https://github.com/NVIDIA/gdrcopy.git /opt/gdrcopy \
+# legacy NVSHMEM/IBGDA build.
-    && make -C /opt/gdrcopy CUDA="${CUDA_HOME}" prefix=/usr/local lib_install \
+RUN set -eux; \
-    && printf '%s\n' /usr/local/lib > /etc/ld.so.conf.d/gdrcopy.conf \
+    if [ "${CUDA_TAG}" = cu129 ]; then \
-    && ldconfig \
+        git clone --depth 1 --branch "v${GDRCOPY_VERSION}" \
-    && test -f /usr/local/include/gdrapi.h \
+            https://github.com/NVIDIA/gdrcopy.git /opt/gdrcopy; \
-    && ldconfig -p | grep -q libgdrapi
+        make -C /opt/gdrcopy CUDA="${CUDA_HOME}" prefix=/usr/local lib_install; \
+        printf '%s\n' /usr/local/lib > /etc/ld.so.conf.d/gdrcopy.conf; \
+        ldconfig; \
+        test -f /usr/local/include/gdrapi.h; \
+        ldconfig -p | grep -q libgdrapi; \
+    fi
 
 RUN --mount=type=cache,id=sgl-deep-ep-pip-${CUDA_TAG}-${PYTHON_TAG}-${ARCHITECTURE},target=/root/.cache/pip \
     set -eux; \
@@ -66,6 +69,10 @@ RUN --mount=type=cache,id=sgl-deep-ep-pip-${CUDA_TAG}-${PYTHON_TAG}-${ARCHITECTU
     "${PYTHON_BIN}" -m pip install --force-reinstall \
         "torch==${TORCH_VERSION}" \
         --index-url "https://download.pytorch.org/whl/${CUDA_TAG}"; \
+    if [ "${CUDA_TAG}" = cu130 ]; then \
+        "${PYTHON_BIN}" -m pip install --force-reinstall --no-deps \
+            "nvidia-nccl-cu13==${NCCL_VERSION}"; \
+    fi; \
     "${PYTHON_BIN}" -m pip install \
         "auditwheel>=6.0" \
         build \