Add sgl-kernel CI test for Blackwell (B200)

.github/workflows/pr-test.yml

@@ -256,6 +256,36 @@ jobs:
 
           echo "All benchmark tests completed!"
 
+  sgl-kernel-b200-test:
+    needs: [check-changes, sgl-kernel-build-wheels]
+    if: needs.check-changes.outputs.sgl_kernel == 'true'
+    runs-on: 4-gpu-b200
+    env:
+      RUNNER_LABELS: 4-gpu-b200
+    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 dependencies
+        run: |
+          CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} IS_BLACKWELL=1 bash scripts/ci/ci_install_dependency.sh
+
+      - name: Run sgl-kernel unit tests on B200
+        timeout-minutes: 30
+        run: |
+          cd sgl-kernel
+          pytest tests/
   # Adding a single CUDA13 smoke test to verify that the kernel builds and runs
   # TODO: Add back this test when it can pass on CI
   # cuda13-kernel-smoke-test:
@@ -931,6 +961,7 @@ jobs:
         sgl-kernel-unit-test,
         sgl-kernel-mla-test,
         sgl-kernel-benchmark-test,
+        sgl-kernel-b200-test,
 
         multimodal-gen-test,
 

sgl-kernel/README.md

@@ -9,7 +9,7 @@
 
 </div>
 
-SGL Kernel provides optimized compute primitives for the SGLang framework, enabling efficient inference for large language models and vision-language models through custom kernels for operations.
+SGL Kernel provides optimized compute primitives for the SGLang framework, enabling efficient inference for large language models and vision-language models through custom kernel operations.
 
 ## Installation
 Requires torch == 2.8.0

sgl-kernel/tests/conftest.py

@@ -2,7 +2,7 @@
 import torch
 
 
-# This ensures the torch defaults don't get left in modified states between
+# This fixture ensures the torch defaults don't get left in modified states between
 # tests (e.g., when a test fails before restoring the original value), which
 # can cause subsequent tests to fail.
 @pytest.fixture(autouse=True)