Adapt fixture-kit to gsm8k mixin (#15599)

test/srt/models/test_ministral3_models.py

@@ -1,20 +1,20 @@
 import unittest
 from types import SimpleNamespace
 
-from sglang.test.gsm8k_mixin import GSM8KMixin
-from sglang.test.mmmu_vlm_mixin import MMMUVLMMixin
-from sglang.test.test_utils import CustomTestCase
+from sglang.test.kits.gsm8k_accuracy_kit import GSM8KMixin
+from sglang.test.kits.mmmu_vlm_kit import MMMUVLMTestBase
+from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
 MODEL = "mistralai/Ministral-3-3B-Instruct-2512"
 
 
-class TestMinistral3TextOnly(GSM8KMixin, CustomTestCase):
+class TestMinistral3TextOnly(GSM8KMixin, DefaultServerBase):
     accuracy = 0.6
     model = MODEL
     other_args = ["--trust-remote-code"]
 
 
-class TestMinistral3MMMU(MMMUVLMMixin, CustomTestCase):
+class TestMinistral3MMMU(MMMUVLMTestBase):
     accuracy = 0.3
     model = MODEL
     other_args = ["--trust-remote-code"]

test/srt/models/test_nvidia_nemotron_nano_v2.py

@@ -1,24 +1,24 @@
 import unittest
 
 from sglang.srt.utils import is_blackwell
-from sglang.test.gsm8k_mixin import GSM8KMixin
-from sglang.test.test_utils import CustomTestCase
+from sglang.test.kits.gsm8k_accuracy_kit import GSM8KMixin
+from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
 
-class TestNvidiaNemotronNanoV2BF16(GSM8KMixin, CustomTestCase):
+class TestNvidiaNemotronNanoV2BF16(GSM8KMixin, DefaultServerBase):
     model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
     accuracy = 0.87
     other_args = ["--max-mamba-cache-size", "256"]
 
 
-class TestNvidiaNemotronNanoV2FP8(GSM8KMixin, CustomTestCase):
+class TestNvidiaNemotronNanoV2FP8(GSM8KMixin, DefaultServerBase):
     accuracy = 0.87
     model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2-FP8"
     other_args = ["--max-mamba-cache-size", "256"]
 
 
 @unittest.skipIf(not is_blackwell(), "NVFP4 only supported on blackwell")
-class TestNvidiaNemotronNanoV2NVFP4(GSM8KMixin, CustomTestCase):
+class TestNvidiaNemotronNanoV2NVFP4(GSM8KMixin, DefaultServerBase):
     accuracy = 0.855
     model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2-NVFP4"
     other_args = ["--max-mamba-cache-size", "256"]
@@ -28,7 +28,7 @@ class TestNvidiaNemotronNanoV2NVFP4(GSM8KMixin, CustomTestCase):
     "STANDALONE speculative decoding does not yet support target and draft models "
     "with different hidden sizes (Nemotron-9B: 4480, Llama-3.2-1B: 2048)"
 )
-class TestNvidiaNemotronNanoV2SpeculativeDecoding(GSM8KMixin, CustomTestCase):
+class TestNvidiaNemotronNanoV2SpeculativeDecoding(GSM8KMixin, DefaultServerBase):
     accuracy = 0.87
     model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
     other_args = [
@@ -57,7 +57,9 @@ class TestNvidiaNemotronNanoV2SpeculativeDecoding(GSM8KMixin, CustomTestCase):
     "STANDALONE speculative decoding does not yet support target and draft models "
     "with different hidden sizes (Nemotron-9B: 4480, Llama-3.2-1B: 2048)"
 )
-class TestNvidiaNemotronNanoV2SpeculativeDecodingBF16Cache(GSM8KMixin, CustomTestCase):
+class TestNvidiaNemotronNanoV2SpeculativeDecodingBF16Cache(
+    GSM8KMixin, DefaultServerBase
+):
     accuracy = 0.87
     model = "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
     other_args = [

test/srt/models/test_nvidia_nemotron_nano_v2_vl.py

@@ -1,20 +1,20 @@
 import unittest
 from types import SimpleNamespace
 
-from sglang.test.gsm8k_mixin import GSM8KMixin
-from sglang.test.mmmu_vlm_mixin import MMMUVLMMixin
-from sglang.test.test_utils import CustomTestCase
+from sglang.test.kits.gsm8k_accuracy_kit import GSM8KMixin
+from sglang.test.kits.mmmu_vlm_kit import MMMUVLMTestBase
+from sglang.test.server_fixtures.default_fixture import DefaultServerBase
 
 MODEL = "nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16"
 
 
-class TestNvidiaNemotronNanoV2VLTextOnly(GSM8KMixin, CustomTestCase):
+class TestNvidiaNemotronNanoV2VLTextOnly(GSM8KMixin, DefaultServerBase):
     accuracy = 0.87
     model = MODEL
     other_args = ["--max-mamba-cache-size", "256", "--trust-remote-code"]
 
 
-class TestNvidiaNemotronNanoV2VLMMMU(MMMUVLMMixin, CustomTestCase):
+class TestNvidiaNemotronNanoV2VLMMMU(MMMUVLMTestBase):
     accuracy = 0.454
     model = MODEL
     other_args = ["--max-mamba-cache-size", "128", "--trust-remote-code"]

test/srt/models/test_vlm_models.py

@@ -5,8 +5,8 @@ import unittest
 from types import SimpleNamespace
 
 from sglang.srt.utils import is_hip
-from sglang.test.mmmu_vlm_mixin import DEFAULT_MEM_FRACTION_STATIC, MMMUVLMMixin
-from sglang.test.test_utils import CustomTestCase, is_in_ci
+from sglang.test.kits.mmmu_vlm_kit import DEFAULT_MEM_FRACTION_STATIC, MMMUVLMTestBase
+from sglang.test.test_utils import is_in_ci
 
 _is_hip = is_hip()
 # VLM models for testing
@@ -20,7 +20,7 @@ else:
     ]
 
 
-class TestVLMModels(MMMUVLMMixin, CustomTestCase):
+class TestVLMModels(MMMUVLMTestBase):
     def test_vlm_mmmu_benchmark(self):
         """Test VLM models against MMMU benchmark."""
         models_to_test = MODELS