Let sgl-kernel changes be tested on srt (#10313)

.github/workflows/pr-test-sgl-kernel.yml

@@ -1,151 +0,0 @@
-name: PR Test (sgl-kernel)
-
-on:
-  push:
-    branches: [main]
-    paths:
-      - "sgl-kernel/**"
-  pull_request:
-    branches: [main]
-    paths:
-      - "sgl-kernel/**"
-  workflow_dispatch:
-
-concurrency:
-  group: pr-test-sgl-kernel-${{ github.ref }}
-  cancel-in-progress: true
-
-jobs:
-  lint:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Check clang-format
-        uses: DoozyX/clang-format-lint-action@v0.18.1
-        with:
-          source: sgl-kernel
-          extensions: h,c,cpp,hpp,cu,cuh,cc
-          clangFormatVersion: 18
-          style: file
-
-  build-wheels:
-    if: github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request'
-    runs-on: sgl-kernel-build-node
-    strategy:
-      matrix:
-        include:
-          - python-version: "3.10"
-            cuda-version: "12.4"
-          - python-version: "3.10"
-            cuda-version: "12.8"
-          - python-version: "3.10"
-            cuda-version: "12.9"
-    name: Build Wheel (CUDA ${{ matrix.cuda-version }})
-    steps:
-      - name: Cleanup
-        run: |
-          sudo rm -rf $GITHUB_WORKSPACE/* || true
-
-      - uses: actions/checkout@v4
-        with:
-          submodules: "recursive"
-
-      - name: Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v5
-        with:
-          python-version: ${{ matrix.python-version }}
-
-      - name: Build wheel for Python ${{ matrix.python-version }} and CUDA ${{ matrix.cuda-version }}
-        if: github.event_name != 'push' || (matrix.cuda-version != '12.4' && matrix.cuda-version != '12.8')
-        run: |
-          cd sgl-kernel
-          chmod +x ./build.sh
-          ./build.sh "${{ matrix.python-version }}" "${{ matrix.cuda-version }}"
-
-      - name: Upload artifacts
-        uses: actions/upload-artifact@v4
-        with:
-          name: wheel-python${{ matrix.python-version }}-cuda${{ matrix.cuda-version }}
-          path: sgl-kernel/dist/*
-
-  unit-test:
-    if: github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request'
-    needs: build-wheels
-    runs-on: 1-gpu-runner
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Download artifacts
-        uses: actions/download-artifact@v4
-        with:
-          path: sgl-kernel/dist/
-          merge-multiple: true
-          pattern: wheel-python3.10-cuda12.9
-
-      - name: Install
-        run: |
-          bash scripts/ci/ci_install_dependency.sh
-          pip3 install torch==2.8.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/cu126 && pip3 install pytest
-          pip3 uninstall sgl-kernel -y || true
-          pip3 install sgl-kernel/dist/*whl --force-reinstall --no-deps
-          pip3 list | grep sgl-kernel
-
-      - name: Run test
-        timeout-minutes: 30
-        run: |
-          cd sgl-kernel
-          pytest tests/
-
-      - name: Uninstall dependencies
-        run: |
-          pip3 uninstall sgl-kernel -y
-
-  mla-test:
-    if: github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request'
-    needs: build-wheels
-    runs-on: 1-gpu-runner
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Download artifacts
-        uses: actions/download-artifact@v4
-        with:
-          path: sgl-kernel/dist/
-          merge-multiple: true
-          pattern: wheel-python3.10-cuda12.9
-
-      - name: Install
-        run: |
-          bash scripts/ci/ci_install_dependency.sh
-          pip3 install torch==2.8.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/cu126
-          pip3 uninstall sgl-kernel -y || true
-          pip3 install sgl-kernel/dist/*whl --force-reinstall --no-deps
-          pip3 list | grep sgl-kernel
-
-      - name: Run test
-        timeout-minutes: 30
-        run: |
-          cd test/srt
-          python3 test_mla_deepseek_v3.py
-
-      - name: Uninstall dependencies
-        run: |
-          pip3 uninstall sgl-kernel -y
-
-  finish:
-    needs: [unit-test, mla-test, lint, build-wheels]
-    runs-on: ubuntu-latest
-    steps:
-      - name: Check all dependent job statuses
-        run: |
-          results=(${{ join(needs.*.result, ' ') }})
-          for result in "${results[@]}"; do
-            if [ "$result" = "failure" ] || [ "$result" = "cancelled" ]; then
-              echo "Job failed with result: $result"
-              exit 1
-            fi
-          done
-          echo "All jobs completed successfully"
-          exit 0

.github/workflows/pr-test.yml

@@ -21,6 +21,182 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
+  # =============================================== sgl-kernel ====================================================
+
+  sgl-kernel-check-changes:
+    runs-on: ubuntu-latest
+    outputs:
+      src: ${{ steps.filter.outputs.src }}
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Detect file changes
+        id: filter
+        uses: dorny/paths-filter@v3
+        with:
+          filters: |
+            src:
+              - "sgl-kernel/**"
+
+  sgl-kernel-lint:
+    runs-on: ubuntu-latest
+    needs: sgl-kernel-check-changes
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
+      github.event.pull_request.draft == false &&
+      needs.sgl-kernel-check-changes.outputs.src == 'true'
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Check clang-format
+        uses: DoozyX/clang-format-lint-action@v0.18.1
+        with:
+          source: sgl-kernel
+          extensions: h,c,cpp,hpp,cu,cuh,cc
+          clangFormatVersion: 18
+          style: file
+
+  sgl-kernel-build-wheels:
+    needs: sgl-kernel-check-changes
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
+      github.event.pull_request.draft == false &&
+      needs.sgl-kernel-check-changes.outputs.src == 'true'
+    runs-on: sgl-kernel-build-node
+    strategy:
+      matrix:
+        include:
+          - python-version: "3.10"
+            cuda-version: "12.4"
+          - python-version: "3.10"
+            cuda-version: "12.8"
+          - python-version: "3.10"
+            cuda-version: "12.9"
+    name: Build Wheel (CUDA ${{ matrix.cuda-version }})
+    steps:
+      - name: Cleanup
+        run: |
+          sudo rm -rf $GITHUB_WORKSPACE/* || true
+
+      - uses: actions/checkout@v4
+        with:
+          submodules: "recursive"
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Build wheel for Python ${{ matrix.python-version }} and CUDA ${{ matrix.cuda-version }}
+        if: github.event_name != 'push' || (matrix.cuda-version != '12.4' && matrix.cuda-version != '12.8')
+        run: |
+          cd sgl-kernel
+          chmod +x ./build.sh
+          ./build.sh "${{ matrix.python-version }}" "${{ matrix.cuda-version }}"
+
+      - name: Upload artifacts
+        uses: actions/upload-artifact@v4
+        with:
+          name: wheel-python${{ matrix.python-version }}-cuda${{ matrix.cuda-version }}
+          path: sgl-kernel/dist/*
+
+  sgl-kernel-unit-test:
+    needs: [sgl-kernel-check-changes, sgl-kernel-build-wheels]
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
+      github.event.pull_request.draft == false &&
+      needs.sgl-kernel-check-changes.outputs.src == 'true'
+    runs-on: 1-gpu-runner
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Cleanup
+        run: |
+          ls -alh sgl-kernel/dist || true
+          rm -rf sgl-kernel/dist/* || true
+
+      - name: Download artifacts
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.9
+
+      - name: Install
+        run: |
+          bash scripts/ci/ci_install_dependency.sh
+          pip3 install torch==2.8.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/cu126 && pip3 install pytest
+          pip3 uninstall sgl-kernel -y || true
+          pip3 install sgl-kernel/dist/*whl --force-reinstall --no-deps
+          pip3 list | grep sgl-kernel
+
+      - name: Run test
+        timeout-minutes: 30
+        run: |
+          cd sgl-kernel
+          pytest tests/
+
+      - name: Uninstall dependencies
+        run: |
+          pip3 uninstall sgl-kernel -y
+
+  sgl-kernel-mla-test:
+    needs: [sgl-kernel-check-changes, sgl-kernel-build-wheels]
+    if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
+      github.event.pull_request.draft == false &&
+      needs.sgl-kernel-check-changes.outputs.src == 'true'
+    runs-on: 1-gpu-runner
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Cleanup
+        run: |
+          ls -alh sgl-kernel/dist || true
+          rm -rf sgl-kernel/dist/* || true
+
+      - name: Download artifacts
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.9
+
+      - name: Install
+        run: |
+          bash scripts/ci/ci_install_dependency.sh
+          pip3 install torch==2.8.0 torchvision torchaudio --index-url https://download.pytorch.org/whl/test/cu126
+          pip3 uninstall sgl-kernel -y || true
+          pip3 install sgl-kernel/dist/*whl --force-reinstall --no-deps
+          pip3 list | grep sgl-kernel
+
+      - name: Run test
+        timeout-minutes: 30
+        run: |
+          cd test/srt
+          python3 test_mla_deepseek_v3.py
+
+      - name: Uninstall dependencies
+        run: |
+          pip3 uninstall sgl-kernel -y
+
+  sgl-kernel-finish:
+    needs: [sgl-kernel-unit-test, sgl-kernel-mla-test, sgl-kernel-lint, sgl-kernel-build-wheels]
+    if: always()
+    runs-on: ubuntu-latest
+    steps:
+      - name: Check all dependent job statuses
+        run: |
+          results=(${{ join(needs.*.result, ' ') }})
+          for result in "${results[@]}"; do
+            if [ "$result" = "failure" ] || [ "$result" = "cancelled" ]; then
+              echo "Job failed with result: $result"
+              exit 1
+            fi
+          done
+          echo "All jobs completed successfully"
+          exit 0
+
+  # =============================================== primary ====================================================
+
   check-changes:
     runs-on: ubuntu-latest
     outputs:
@@ -39,9 +215,10 @@ jobs:
               - "scripts/ci/**"
               - "test/**"
               - ".github/workflows/pr-test.yml"
+              - "sgl-kernel/**"
 
   unit-test-frontend:
-    needs: check-changes
+    needs: [check-changes, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -50,9 +227,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_dependency.sh
 
       - name: Run test
         timeout-minutes: 10
@@ -61,7 +246,7 @@ jobs:
           python3 run_suite.py --suite per-commit
 
   unit-test-backend-1-gpu:
-    needs: [check-changes, unit-test-frontend]
+    needs: [check-changes, unit-test-frontend, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -74,9 +259,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_dependency.sh
 
       - name: Run test
         timeout-minutes: 30
@@ -85,7 +278,7 @@ jobs:
           python3 run_suite.py --suite per-commit --auto-partition-id ${{ matrix.part }} --auto-partition-size 10
 
   unit-test-backend-2-gpu:
-    needs: [check-changes]
+    needs: [check-changes, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -98,9 +291,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_dependency.sh
 
       - name: Run test
         timeout-minutes: 30
@@ -109,7 +310,7 @@ jobs:
           python3 run_suite.py --suite per-commit-2-gpu --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
 
   unit-test-backend-4-gpu:
-    needs: [check-changes, unit-test-backend-2-gpu]
+    needs: [check-changes, unit-test-backend-2-gpu, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -122,9 +323,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_dependency.sh
 
       - name: Run test
         timeout-minutes: 20
@@ -133,7 +342,7 @@ jobs:
           python3 run_suite.py --suite per-commit-4-gpu --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
 
   unit-test-backend-8-gpu:
-    needs: [check-changes, unit-test-backend-2-gpu]
+    needs: [check-changes, unit-test-backend-2-gpu, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -146,9 +355,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_dependency.sh
 
       - name: Run test
         timeout-minutes: 20
@@ -157,7 +374,7 @@ jobs:
           python3 run_suite.py --suite per-commit-8-gpu --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
 
   performance-test-1-gpu-part-1:
-    needs: check-changes
+    needs: [check-changes, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -166,9 +383,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_dependency.sh
 
       - name: Benchmark single latency
         timeout-minutes: 10
@@ -209,7 +434,7 @@ jobs:
           python3 -m unittest test_bench_serving.TestBenchServing.test_lora_online_latency_with_concurrent_adapter_updates
 
   performance-test-1-gpu-part-2:
-    needs: check-changes
+    needs: [check-changes, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -218,9 +443,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_dependency.sh
 
       - name: Benchmark offline throughput (w/o RadixAttention)
         timeout-minutes: 10
@@ -253,7 +486,7 @@ jobs:
           python3 -m unittest test_bench_serving.TestBenchServing.test_vlm_online_latency
 
   performance-test-2-gpu:
-    needs: [check-changes, unit-test-backend-2-gpu]
+    needs: [check-changes, unit-test-backend-2-gpu, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -262,9 +495,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_dependency.sh
 
       - name: Benchmark single latency (TP=2)
         timeout-minutes: 10
@@ -303,7 +544,7 @@ jobs:
           python3 -m unittest test_bench_serving.TestBenchServing.test_pp_long_context_prefill
 
   accuracy-test-1-gpu:
-    needs: check-changes
+    needs: [check-changes, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -312,9 +553,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_dependency.sh
           git clone https://github.com/merrymercy/human-eval.git
           cd human-eval
           pip install -e .
@@ -326,7 +575,7 @@ jobs:
           python3 test_eval_accuracy_large.py
 
   accuracy-test-2-gpu:
-    needs: [check-changes, accuracy-test-1-gpu]
+    needs: [check-changes, accuracy-test-1-gpu, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -335,9 +584,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_dependency.sh
           git clone https://github.com/merrymercy/human-eval.git
           cd human-eval
           pip install -e .
@@ -349,7 +606,7 @@ jobs:
           python3 test_moe_eval_accuracy_large.py
 
   unit-test-deepep-4-gpu:
-    needs: [check-changes, unit-test-backend-2-gpu]
+    needs: [check-changes, unit-test-backend-2-gpu, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -358,9 +615,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_deepep.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_deepep.sh
 
       - name: Run test
         timeout-minutes: 20
@@ -369,7 +634,7 @@ jobs:
           python3 run_suite.py --suite per-commit-4-gpu-deepep
 
   unit-test-deepep-8-gpu:
-    needs: [check-changes, unit-test-backend-2-gpu]
+    needs: [check-changes, unit-test-backend-2-gpu, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
         github.event.pull_request.draft == false &&
         needs.check-changes.outputs.src == 'true'
@@ -378,9 +643,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.4
+
       - name: Install dependencies
         run: |
-          bash scripts/ci/ci_install_deepep.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} bash scripts/ci/ci_install_deepep.sh
 
       - name: Run test
         timeout-minutes: 20
@@ -389,7 +662,7 @@ jobs:
           python3 run_suite.py --suite per-commit-8-gpu-deepep
 
   unit-test-backend-8-gpu-b200:
-    needs: [check-changes, unit-test-backend-2-gpu]
+    needs: [check-changes, unit-test-backend-2-gpu, sgl-kernel-check-changes, sgl-kernel-finish]
     if: (github.repository == 'sgl-project/sglang' || github.event_name == 'pull_request') &&
       github.event.pull_request.draft == false &&
       needs.check-changes.outputs.src == 'true'
@@ -400,9 +673,17 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
+      - name: Download artifacts
+        if: needs.check-changes.outputs.src == 'true'
+        uses: actions/download-artifact@v4
+        with:
+          path: sgl-kernel/dist/
+          merge-multiple: true
+          pattern: wheel-python3.10-cuda12.8
+
       - name: Install dependencies
         run: |
-          IS_BLACKWELL=1 bash scripts/ci/ci_install_dependency.sh
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.sgl-kernel-check-changes.outputs.src}} IS_BLACKWELL=1 bash scripts/ci/ci_install_dependency.sh
 
       - name: Run test
         timeout-minutes: 60

scripts/ci/ci_install_dependency.sh

@@ -53,9 +53,16 @@ SGL_KERNEL_VERSION_FROM_SRT=$(grep -Po -m1 '(?<=sgl-kernel==)[0-9A-Za-z\.\-]+' p
 echo "SGL_KERNEL_VERSION_FROM_KERNEL=${SGL_KERNEL_VERSION_FROM_KERNEL} SGL_KERNEL_VERSION_FROM_SRT=${SGL_KERNEL_VERSION_FROM_SRT}"
 
 if [ "$IS_BLACKWELL" = "1" ]; then
-    $PIP_CMD install https://github.com/sgl-project/whl/releases/download/v${SGL_KERNEL_VERSION_FROM_SRT}/sgl_kernel-${SGL_KERNEL_VERSION_FROM_SRT}+cu128-cp310-abi3-manylinux2014_x86_64.whl --force-reinstall $PIP_INSTALL_SUFFIX
+    SGL_KERNEL_CUDA_VERSION=cu128
 else
-    $PIP_CMD install https://github.com/sgl-project/whl/releases/download/v${SGL_KERNEL_VERSION_FROM_SRT}/sgl_kernel-${SGL_KERNEL_VERSION_FROM_SRT}+cu124-cp310-abi3-manylinux2014_x86_64.whl --force-reinstall $PIP_INSTALL_SUFFIX
+    SGL_KERNEL_CUDA_VERSION=cu124
+fi
+
+if [ "${CUSTOM_BUILD_SGL_KERNEL:-}" = "true" ]; then
+    ls -alh sgl-kernel/dist
+    $PIP_CMD install sgl-kernel/dist/sgl_kernel-${SGL_KERNEL_VERSION_FROM_KERNEL}+${SGL_KERNEL_CUDA_VERSION}-cp310-abi3-manylinux2014_x86_64.whl --force-reinstall $PIP_INSTALL_SUFFIX
+else
+    $PIP_CMD install https://github.com/sgl-project/whl/releases/download/v${SGL_KERNEL_VERSION_FROM_SRT}/sgl_kernel-${SGL_KERNEL_VERSION_FROM_SRT}+${SGL_KERNEL_CUDA_VERSION}-cp310-abi3-manylinux2014_x86_64.whl --force-reinstall $PIP_INSTALL_SUFFIX
 fi
 
 # Show current packages