[CI] Migrate nightly tests to test/registered/ (#15582)

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

@@ -191,7 +191,7 @@ jobs:
         timeout-minutes: 120
         run: |
           cd test
-          python3 nightly/test_text_models_gsm8k_eval.py
+          python3 run_suite.py --hw cuda --suite nightly-eval-text-2-gpu --nightly --continue-on-error --timeout-per-file 4500
 
   # Text model performance tests
   nightly-test-text-perf-2-gpu-runner:
@@ -216,7 +216,7 @@ jobs:
         run: |
           cd test
           rm -rf performance_profiles_text_models/
-          python3 nightly/test_text_models_perf.py
+          python3 run_suite.py --hw cuda --suite nightly-perf-text-2-gpu --nightly --continue-on-error
 
       - name: Publish traces to storage repo
         env:
@@ -244,7 +244,7 @@ jobs:
         timeout-minutes: 240
         run: |
           cd test
-          python3 nightly/test_vlms_mmmu_eval.py
+          python3 run_suite.py --hw cuda --suite nightly-eval-vlm-2-gpu --nightly --continue-on-error --timeout-per-file 9000
 
   # VLM performance tests
   nightly-test-vlm-perf-2-gpu-runner:
@@ -269,7 +269,7 @@ jobs:
         run: |
           cd test
           rm -rf performance_profiles_vlms/
-          python3 nightly/test_vlms_perf.py
+          python3 run_suite.py --hw cuda --suite nightly-perf-vlm-2-gpu --nightly --continue-on-error
 
       - name: Publish traces to storage repo
         env:

python/sglang/test/ascend/vlm_utils.py

@@ -87,7 +87,7 @@ class TestVLMModels(CustomTestCase):
             "--limit",
             limit,
             "--config",
-            "/__w/sglang/sglang/test/nightly/ascend/vlm_models/mmmu-val.yaml",
+            "/__w/sglang/sglang/test/registered/ascend/vlm_models/mmmu-val.yaml",
         ]
 
         subprocess.run(

python/sglang/test/performance_test_runner.py

@@ -1,9 +1,8 @@
 from dataclasses import dataclass, field
 from typing import List, Optional, Tuple
 
-from nightly_utils import NightlyBenchmarkRunner
-
 from sglang.test.nightly_bench_utils import BenchmarkResult
+from sglang.test.nightly_utils import NightlyBenchmarkRunner
 from sglang.test.test_utils import DEFAULT_URL_FOR_TEST, ModelLaunchSettings
 
 

python/sglang/test/run_combined_tests.py

@@ -1,18 +1,17 @@
 from typing import List, Optional
 
-from accuracy_test_runner import (
+from sglang.test.accuracy_test_runner import (
     AccuracyTestParams,
     AccuracyTestResult,
     run_accuracy_test,
     write_accuracy_github_summary,
 )
-from nightly_utils import NightlyBenchmarkRunner
+from sglang.test.nightly_utils import NightlyBenchmarkRunner
-from performance_test_runner import (
+from sglang.test.performance_test_runner import (
     PerformanceTestParams,
     PerformanceTestResult,
     run_performance_test,
 )
-
 from sglang.test.test_utils import DEFAULT_URL_FOR_TEST, ModelLaunchSettings, is_in_ci
 
 

test/registered/8-gpu-models/test_deepseek_v31.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, is_blackwell_system
 
 # Runs on both H200 and B200 via nightly-8-gpu-common suite

test/registered/8-gpu-models/test_deepseek_v32.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, is_blackwell_system
 
 register_cuda_ci(est_time=8000, suite="nightly-8-gpu-common", nightly=True)

test/registered/8-gpu-models/test_glm_46.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

test/registered/8-gpu-models/test_kimi_k2.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

test/registered/8-gpu-models/test_minimax_m2.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

test/registered/8-gpu-models/test_mistral_large3.py

@@ -1,16 +1,10 @@
 import os
-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, is_blackwell_system
 
 # Runs on both H200 and B200 via nightly-8-gpu-common suite

test/registered/8-gpu-models/test_qwen3_235b.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, is_blackwell_system
 
 # Runs on both H200 and B200 via nightly-8-gpu-common suite

test/registered/ascend/llm_models/test_ascend_afm_4_5b.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_baichuan2_13b_chat.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_charglm2_6b.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_exaone_3.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_gemma_3_1b_it.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_glm4_9b_chat.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_granite_3_0_3b_a800m.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_granite_3_1_8b.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_internlm2_7b.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_ling_lite.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_llama_2_7b.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_mimo_7b_rl.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_mistral_7b.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_persimmon_8b_chat.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_phi_4_multimodal.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/llm_models/test_ascend_smollm_1_7b.py

@@ -1,7 +1,6 @@
 import unittest
 
-from gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
-
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 

test/registered/ascend/vlm_models/test_ascend_gemma_3_4b_it.py

@@ -1,7 +1,6 @@
 import unittest
 
-from test_vlm_utils import TestVLMModels
+from sglang.test.ascend.vlm_utils import TestVLMModels
-
 from sglang.test.ci.ci_register import register_npu_ci
 
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)

test/registered/ascend/vlm_models/test_ascend_janus_pro_1b.py

@@ -1,7 +1,6 @@
 import unittest
 
-from test_vlm_utils import TestVLMModels
+from sglang.test.ascend.vlm_utils import TestVLMModels
-
 from sglang.test.ci.ci_register import register_npu_ci
 
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)

test/registered/ascend/vlm_models/test_ascend_janus_pro_7b.py

@@ -1,7 +1,6 @@
 import unittest
 
-from test_vlm_utils import TestVLMModels
+from sglang.test.ascend.vlm_utils import TestVLMModels
-
 from sglang.test.ci.ci_register import register_npu_ci
 
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)

test/registered/ascend/vlm_models/test_ascend_mimo_vl_7b_rl.py

@@ -1,7 +1,6 @@
 import unittest
 
-from test_vlm_utils import TestVLMModels
+from sglang.test.ascend.vlm_utils import TestVLMModels
-
 from sglang.test.ci.ci_register import register_npu_ci
 
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)

test/registered/ascend/vlm_models/test_ascend_minicpm_o_2_6.py

@@ -1,7 +1,6 @@
 import unittest
 
-from test_vlm_utils import TestVLMModels
+from sglang.test.ascend.vlm_utils import TestVLMModels
-
 from sglang.test.ci.ci_register import register_npu_ci
 
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)

test/registered/ascend/vlm_models/test_ascend_minicpm_v_2_6.py

@@ -1,7 +1,6 @@
 import unittest
 
-from test_vlm_utils import TestVLMModels
+from sglang.test.ascend.vlm_utils import TestVLMModels
-
 from sglang.test.ci.ci_register import register_npu_ci
 
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)

test/registered/ascend/vlm_models/test_ascend_phi4_multimodal_instruct.py

@@ -1,7 +1,6 @@
 import unittest
 
-from test_vlm_utils import TestVLMModels
+from sglang.test.ascend.vlm_utils import TestVLMModels
-
 from sglang.test.ci.ci_register import register_npu_ci
 
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)

test/registered/ascend/vlm_models/test_ascend_qwen2_5_vl_3b_instruct.py

@@ -1,7 +1,6 @@
 import unittest
 
-from test_vlm_utils import TestVLMModels
+from sglang.test.ascend.vlm_utils import TestVLMModels
-
 from sglang.test.ci.ci_register import register_npu_ci
 
 register_npu_ci(est_time=400, suite="nightly-4-npu-a3", nightly=True)

test/registered/backends/test_deepseek_v3_fp4_cutlass_moe.py

@@ -0,0 +1,75 @@
+import unittest
+from types import SimpleNamespace
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
+from sglang.test.test_utils import (
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    is_in_ci,
+    popen_launch_server,
+    write_github_step_summary,
+)
+
+register_cuda_ci(est_time=900, suite="nightly-4-gpu-b200", nightly=True)
+
+FULL_DEEPSEEK_V3_FP4_MODEL_PATH = "nvidia/DeepSeek-V3-0324-FP4"
+SERVER_LAUNCH_TIMEOUT = 1000
+
+
+class TestDeepseekV3FP4CutlassMoE(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = FULL_DEEPSEEK_V3_FP4_MODEL_PATH
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        other_args = [
+            "--tp",
+            "4",
+            "--ep",
+            "4",
+            "--attention-backend",
+            "trtllm_mla",
+            "--moe-runner-backend",
+            "flashinfer_cutlass",
+            "--quantization",
+            "modelopt_fp4",
+            "--model-loader-extra-config",
+            '{"enable_multithread_load": true}',
+        ]
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=SERVER_LAUNCH_TIMEOUT,
+            other_args=other_args,
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_a_gsm8k(
+        self,
+    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
+        args = SimpleNamespace(
+            num_shots=8,
+            data_path=None,
+            num_questions=1319,
+            parallel=1319,
+            max_new_tokens=512,
+            host="http://127.0.0.1",
+            port=int(self.base_url.split(":")[-1]),
+        )
+        metrics = run_eval_few_shot_gsm8k(args)
+        print(f"{metrics=}")
+
+        if is_in_ci():
+            write_github_step_summary(
+                f"### test_gsm8k (deepseek-v3-fp4-cutlass-moe)\n"
+                f'{metrics["accuracy"]=:.3f}\n'
+            )
+            self.assertGreater(metrics["accuracy"], 0.935)
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/backends/test_qwen3_fp4_trtllm_gen_moe.py

@@ -1,7 +1,8 @@
 import unittest
 from types import SimpleNamespace
 
-from sglang.srt.utils import kill_process_tree
+from sglang.srt.utils import get_device_sm, kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -10,7 +11,13 @@ from sglang.test.test_utils import (
     popen_launch_server,
 )
 
+# modelopt_fp4 requires SM 100+ (Blackwell)
+register_cuda_ci(est_time=300, suite="nightly-1-gpu", nightly=True)
 
+
+@unittest.skipIf(
+    get_device_sm() < 100, "Test requires CUDA SM 100 or higher (Blackwell)"
+)
 class TestFlashinferTrtllmGenMoeBackend(CustomTestCase):
     @classmethod
     def setUpClass(cls):

test/registered/eval/test_text_models_gsm8k_eval.py

@@ -4,6 +4,7 @@ import warnings
 from types import SimpleNamespace
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_MODEL_NAME_FOR_NIGHTLY_EVAL_FP8_TP1,
@@ -19,6 +20,8 @@ from sglang.test.test_utils import (
     write_results_to_json,
 )
 
+register_cuda_ci(est_time=3600, suite="nightly-eval-text-2-gpu", nightly=True)
+
 MODEL_SCORE_THRESHOLDS = {
     "meta-llama/Llama-3.1-8B-Instruct": 0.82,
     "mistralai/Mistral-7B-Instruct-v0.3": 0.58,

test/registered/eval/test_vlms_mmmu_eval.py

@@ -4,6 +4,7 @@ import warnings
 from types import SimpleNamespace
 
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.run_eval import run_eval
 from sglang.test.test_utils import (
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
@@ -15,6 +16,8 @@ from sglang.test.test_utils import (
     write_results_to_json,
 )
 
+register_cuda_ci(est_time=7200, suite="nightly-eval-vlm-2-gpu", nightly=True)
+
 MODEL_THRESHOLDS = {
     # Conservative thresholds on 100 MMMU samples, especially for latency thresholds
     ModelLaunchSettings("deepseek-ai/deepseek-vl2-small"): ModelEvalMetrics(

test/registered/perf/test_gpt_oss_4gpu_perf.py

@@ -1,8 +1,7 @@
 import unittest
 
-from nightly_utils import NightlyBenchmarkRunner
-
 from sglang.test.ci.ci_register import register_cuda_ci
+from sglang.test.nightly_utils import NightlyBenchmarkRunner
 from sglang.test.test_utils import DEFAULT_URL_FOR_TEST
 
 register_cuda_ci(est_time=600, suite="nightly-4-gpu-b200", nightly=True)
@@ -38,7 +37,7 @@ class TestNightlyGptOss4GpuPerformance(unittest.TestCase):
 
         for model_path, other_args in self.models:
             with self.subTest(model=model_path):
-                results, success = self.runner.run_benchmark_for_model(
+                results, success, _ = self.runner.run_benchmark_for_model(
                     model_path=model_path,
                     batch_sizes=self.batch_sizes,
                     input_lens=self.input_lens,

test/registered/perf/test_text_models_perf.py

@@ -1,7 +1,7 @@
 import unittest
 
-from nightly_utils import NightlyBenchmarkRunner
+from sglang.test.ci.ci_register import register_cuda_ci
-
+from sglang.test.nightly_utils import NightlyBenchmarkRunner
 from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     ModelLaunchSettings,
@@ -9,6 +9,8 @@ from sglang.test.test_utils import (
     parse_models,
 )
 
+register_cuda_ci(est_time=3600, suite="nightly-perf-text-2-gpu", nightly=True)
+
 PROFILE_DIR = "performance_profiles_text_models"
 
 
@@ -37,7 +39,7 @@ class TestNightlyTextModelsPerformance(unittest.TestCase):
 
         for model_setup in self.models:
             with self.subTest(model=model_setup.model_path):
-                results, success = self.runner.run_benchmark_for_model(
+                results, success, _ = self.runner.run_benchmark_for_model(
                     model_path=model_setup.model_path,
                     batch_sizes=self.batch_sizes,
                     input_lens=self.input_lens,

test/registered/perf/test_vlms_perf.py

@@ -2,8 +2,8 @@ import os
 import unittest
 import warnings
 
-from nightly_utils import NightlyBenchmarkRunner
+from sglang.test.ci.ci_register import register_cuda_ci
-
+from sglang.test.nightly_utils import NightlyBenchmarkRunner
 from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     ModelLaunchSettings,
@@ -11,6 +11,8 @@ from sglang.test.test_utils import (
     parse_models,
 )
 
+register_cuda_ci(est_time=7200, suite="nightly-perf-vlm-2-gpu", nightly=True)
+
 PROFILE_DIR = "performance_profiles_vlms"
 
 MODEL_DEFAULTS = [
@@ -64,7 +66,7 @@ class TestNightlyVLMModelsPerformance(unittest.TestCase):
                     "--dataset-name=mmmu",
                 ]
 
-                results, success = self.runner.run_benchmark_for_model(
+                results, success, _ = self.runner.run_benchmark_for_model(
                     model_path=model_setup.model_path,
                     batch_sizes=self.batch_sizes,
                     input_lens=self.input_lens,

test/registered/spec/eagle/test_eagle_infer_beta_dp_attention_large.py

@@ -5,6 +5,7 @@ import requests
 
 from sglang.srt.environ import envs
 from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.test_utils import (
     DEFAULT_DEEPSEEK_NVFP4_MODEL_FOR_TEST,
@@ -16,6 +17,9 @@ from sglang.test.test_utils import (
     write_github_step_summary,
 )
 
+# 16 GPU test (4 TP x 4 DP), runs on 2x 8-GPU B200 nodes
+register_cuda_ci(est_time=600, suite="nightly-8-gpu-b200", nightly=True)
+
 
 def test_gsm8k(base_url: str):
     requests.get(base_url + "/flush_cache")

test/run_suite.py

@@ -43,6 +43,11 @@ NIGHTLY_SUITES = {
         "nightly-8-gpu-h200-basic",  # Basic tests for large models on H200
         "nightly-8-gpu-b200-basic",  # Basic tests for large models on B200
         "nightly-8-gpu-common",  # Common tests that run on both H200 and B200
+        # Eval and perf suites (2-gpu)
+        "nightly-eval-text-2-gpu",
+        "nightly-eval-vlm-2-gpu",
+        "nightly-perf-text-2-gpu",
+        "nightly-perf-vlm-2-gpu",
     ],
     HWBackend.AMD: ["nightly-amd", "nightly-amd-8-gpu"],
     HWBackend.CPU: [],
@@ -149,11 +154,7 @@ def run_a_suite(args):
     auto_partition_id = args.auto_partition_id
     auto_partition_size = args.auto_partition_size
 
-    # Temporary: search broadly for nightly tests during migration to registered/
+    # All tests (per-commit and nightly) are now in registered/
-    if nightly:
-        files = glob.glob("**/*.py", recursive=True)
-        sanity_check = False  # Allow files without registration during migration
-    else:
     files = glob.glob("registered/**/*.py", recursive=True)
     # Strict: all registered files must have proper registration
     sanity_check = True