[NVIDIA] Fix broken fp8 MoE of deepseek v3 (#13264)

Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
Co-authored-by: Kangyan-Zhou <zky314343421@gmail.com>

.github/workflows/nightly-test.yml

@@ -182,6 +182,27 @@ jobs:
           cd test/srt
           python3 run_suite.py --suite nightly-8-gpu-h20 --continue-on-error
 
+  nightly-test-8-gpu-b200:
+    if: github.repository == 'sgl-project/sglang'
+    runs-on: 8-gpu-b200
+    env:
+      RUNNER_LABELS: 8-gpu-b200
+    strategy:
+      fail-fast: false
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Install dependencies
+        run: |
+          IS_BLACKWELL=1 bash scripts/ci/ci_install_dependency.sh
+
+      - name: Run test
+        timeout-minutes: 45
+        run: |
+          cd test/srt
+          python3 run_suite.py --suite nightly-8-gpu-b200 --auto-partition-id 0 --auto-partition-size 1 --timeout-per-file 3600
+
   check-all-jobs:
     if: always() && (github.repository == 'sgl-project/sglang' || github.event_name == 'workflow_dispatch')
     needs:
@@ -193,6 +214,7 @@ jobs:
       - nightly-test-4-gpu
       - nightly-test-8-gpu-h200
       - nightly-test-8-gpu-h20
+      - nightly-test-8-gpu-b200
     runs-on: ubuntu-latest
     steps:
       - name: Check if any job failed