fix: nightly fix b200 gpqa (#15745)

.github/workflows/nightly-test-nvidia.yml

@@ -389,7 +389,6 @@ jobs:
           TRACE_BASE_URL: https://raw.githubusercontent.com/sglang-bot/sglang-ci-data/main/traces/${{ github.run_id }}
           PERFETTO_RELAY_URL: ${{ vars.PERFETTO_RELAY_URL }}
           GPU_CONFIG: "8-gpu-b200"
-          SGLANG_ENABLE_JIT_DEEPGEMM: "0"
         run: |
           cd test
           IS_BLACKWELL=1 python3 run_suite.py --hw cuda --suite nightly-8-gpu-common --nightly --timeout-per-file=12000 --continue-on-error

python/sglang/test/accuracy_test_runner.py

@@ -242,7 +242,12 @@ def run_accuracy_test(
         )
 
     # Validate against baseline
-    score = metrics.get("score", 0.0)
+    # Handle different metric key names: "score", "mean_score" (for GPQA with repeat), "accuracy"
+    score = (
+        metrics.get("score")
+        or metrics.get("mean_score")
+        or metrics.get("accuracy", 0.0)
+    )
     passed = score >= params.baseline_accuracy
     latency = metrics.get("latency")
 

test/registered/8-gpu-models/test_glm_46_fp8.py

@@ -1,15 +1,9 @@
-import sys
 import unittest
-from pathlib import Path
-
-# Add nightly directory to path for run_combined_tests import
-sys.path.insert(0, str(Path(__file__).parent.parent.parent / "nightly"))
-
-from accuracy_test_runner import AccuracyTestParams
-from performance_test_runner import PerformanceTestParams
-from run_combined_tests import run_combined_tests
 
+from sglang.test.accuracy_test_runner import AccuracyTestParams
 from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.performance_test_runner import PerformanceTestParams
+from sglang.test.run_combined_tests import run_combined_tests
 from sglang.test.test_utils import ModelLaunchSettings
 
 # Runs on both H200 and B200 via nightly-8-gpu-common suite