From 677930c2760aea06cfba63b5f43464a60ed32906 Mon Sep 17 00:00:00 2001
From: Qiaolin Yu <liin1211@outlook.com>
Date: Sun, 21 Dec 2025 22:27:02 -0800
Subject: [PATCH] Add glm-4.6-fp8 with/without mtp in nightly ci (#15566)

---
 .../8-gpu-models/test_glm_46_fp8.py           | 67 +++++++++++++++++++
 1 file changed, 67 insertions(+)
 create mode 100644 test/registered/8-gpu-models/test_glm_46_fp8.py

diff --git a/test/registered/8-gpu-models/test_glm_46_fp8.py b/test/registered/8-gpu-models/test_glm_46_fp8.py
new file mode 100644
index 000000000..6eafed22c
--- /dev/null
+++ b/test/registered/8-gpu-models/test_glm_46_fp8.py
@@ -0,0 +1,67 @@
+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.ci.ci_register import register_cuda_ci
+from sglang.test.test_utils import ModelLaunchSettings
+
+# Runs on both H200 and B200 via nightly-8-gpu-common suite
+register_cuda_ci(est_time=12000, suite="nightly-8-gpu-common", nightly=True)
+
+GLM_4_6_FP8_MODEL_PATH = "zai-org/GLM-4.6-FP8"
+
+
+class TestGLM46FP8Unified(unittest.TestCase):
+    """Unified test class for GLM-4.6-FP8 performance and accuracy.
+
+    Single variant with simple TP=8 configuration.
+    Runs BOTH:
+    - Performance test (using NightlyBenchmarkRunner)
+    - Accuracy test (using run_eval with mgsm_en)
+    """
+
+    def test_glm_46_fp8_all_variants(self):
+        """Run performance and accuracy for GLM-4.6-FP8."""
+        base_args = [
+            "--tp=8",
+            "--trust-remote-code",
+        ]
+        mtp_args = [
+            "--speculative-algorithm=EAGLE",
+            "--speculative-num-steps=3",
+            "--speculative-eagle-topk=1",
+            "--speculative-num-draft-tokens=4",
+        ]
+
+        variants = [
+            ModelLaunchSettings(
+                GLM_4_6_FP8_MODEL_PATH,
+                tp_size=8,
+                extra_args=base_args,
+            ),
+            ModelLaunchSettings(
+                GLM_4_6_FP8_MODEL_PATH,
+                tp_size=8,
+                extra_args=base_args + mtp_args,
+            ),
+        ]
+
+        run_combined_tests(
+            models=variants,
+            test_name="GLM-4.6-FP8 Unified",
+            accuracy_params=AccuracyTestParams(dataset="gsm8k", baseline_accuracy=0.80),
+            performance_params=PerformanceTestParams(
+                profile_dir="performance_profiles_glm_4_6_fp8",
+            ),
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()