releaes and fix

.github/workflows/release-docker.yml

@@ -1,4 +1,15 @@
 name: Release Docker Images
+#
+# This workflow builds and publishes both framework and runtime Docker images:
+#
+# Framework images (full development environment):
+#   - lmsysorg/sglang:v{version}, lmsysorg/sglang:latest
+#   - lmsysorg/sglang:v{version}-cu129-{amd64,arm64}
+#
+# Runtime images (production-optimized, ~50% smaller):
+#   - lmsysorg/sglang:v{version}-runtime, lmsysorg/sglang:latest-runtime
+#   - lmsysorg/sglang:v{version}-cu129-{amd64,arm64}-runtime
+#
 on:
   push:
     branches:
@@ -45,12 +56,30 @@ jobs:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
 
-      - name: Build and Push AMD64
+      - name: Build and Push AMD64 Framework
         run: |
           version=$(cat python/sglang/version.py | cut -d'"' -f2)
           tag=v${version}-cu129-amd64
 
           docker buildx build \
+            --target framework \
+            --platform linux/amd64 \
+            --push \
+            -f docker/Dockerfile \
+            --build-arg CUDA_VERSION=${{ matrix.variant.cuda_version }} \
+            --build-arg BUILD_TYPE=${{ matrix.variant.build_type }} \
+            --build-arg GRACE_BLACKWELL=${{ matrix.variant.grace_blackwell }} \
+            -t lmsysorg/sglang:${tag} \
+            --no-cache \
+            .
+
+      - name: Build and Push AMD64 Runtime
+        run: |
+          version=$(cat python/sglang/version.py | cut -d'"' -f2)
+          tag=v${version}-cu129-amd64-runtime
+
+          docker buildx build \
+            --target runtime \
             --platform linux/amd64 \
             --push \
             -f docker/Dockerfile \
@@ -87,12 +116,30 @@ jobs:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
 
-      - name: Build and Push ARM64
+      - name: Build and Push ARM64 Framework
         run: |
           version=$(cat python/sglang/version.py | cut -d'"' -f2)
           tag=v${version}-cu129-arm64
 
           docker buildx build \
+            --target framework \
+            --platform linux/arm64 \
+            --push \
+            -f docker/Dockerfile \
+            --build-arg CUDA_VERSION=${{ matrix.variant.cuda_version }} \
+            --build-arg BUILD_TYPE=${{ matrix.variant.build_type }} \
+            --build-arg GRACE_BLACKWELL=${{ matrix.variant.grace_blackwell }} \
+            -t lmsysorg/sglang:${tag} \
+            --no-cache \
+            .
+
+      - name: Build and Push ARM64 Runtime
+        run: |
+          version=$(cat python/sglang/version.py | cut -d'"' -f2)
+          tag=v${version}-cu129-arm64-runtime
+
+          docker buildx build \
+            --target runtime \
             --platform linux/arm64 \
             --push \
             -f docker/Dockerfile \
@@ -125,14 +172,26 @@ jobs:
         run: |
           version=$(cat python/sglang/version.py | cut -d'"' -f2)
 
-          # Create versioned manifest
+          # Create versioned framework manifest (default)
           docker buildx imagetools create \
             -t lmsysorg/sglang:v${version} \
             lmsysorg/sglang:v${version}-cu129-amd64 \
             lmsysorg/sglang:v${version}-cu129-arm64
 
-          # Create latest manifest
+          # Create latest framework manifest (default)
           docker buildx imagetools create \
             -t lmsysorg/sglang:latest \
             lmsysorg/sglang:v${version}-cu129-amd64 \
             lmsysorg/sglang:v${version}-cu129-arm64
+
+          # Create versioned runtime manifest
+          docker buildx imagetools create \
+            -t lmsysorg/sglang:v${version}-runtime \
+            lmsysorg/sglang:v${version}-cu129-amd64-runtime \
+            lmsysorg/sglang:v${version}-cu129-arm64-runtime
+
+          # Create latest runtime manifest
+          docker buildx imagetools create \
+            -t lmsysorg/sglang:latest-runtime \
+            lmsysorg/sglang:v${version}-cu129-amd64-runtime \
+            lmsysorg/sglang:v${version}-cu129-arm64-runtime

docker/Dockerfile

@@ -460,7 +460,7 @@ FROM nvidia/cuda:${CUDA_VERSION}-cudnn-runtime-ubuntu22.04 AS runtime
 
 ARG CUDA_VERSION
 ARG TARGETARCH
-ARG GDRCOPY_VERSION
+ARG GDRCOPY_VERSION=2.5.1
 
 ENV DEBIAN_FRONTEND=noninteractive \
     CUDA_HOME=/usr/local/cuda \
@@ -538,7 +538,7 @@ COPY --from=framework /sgl-workspace /sgl-workspace
 # Copy GDRCopy runtime libraries (but not the build artifacts)
 COPY --from=framework /usr/lib/libgdrapi.so* /usr/lib/
 COPY --from=framework /usr/bin/gdrcopy_* /usr/bin/
-COPY --from=framework ${GDRCOPY_HOME} ${GDRCOPY_HOME}
+COPY --from=framework /usr/src/gdrdrv-2.5.1 /usr/src/gdrdrv-2.5.1
 
 # Copy essential binaries
 COPY --from=framework /usr/local/bin/diff-so-fancy /usr/local/bin/