[NPU] Add GitHub test summary and deduplicate test code. Part 1 (#23835)

Co-authored-by: Elizaveta Martirosian <elizaveta.martirosian@gmail.com>
Co-authored-by: root <root@localhost.localdomain>
Co-authored-by: Elizaveta Martirosian <you@example.com>
Co-authored-by: ronnie_zheng <zl19940307@163.com>

test/registered/ascend/basic_function/parallel_strategy/expert_parallelism/test_npu_deepep_auto_deepseek_v3_2_w8a8.py

@@ -1,22 +1,16 @@
 import os
 import unittest
-from types import SimpleNamespace
 
-from sglang.srt.utils import kill_process_tree
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
 from sglang.test.ascend.test_ascend_utils import DEEPSEEK_V3_2_W8A8_WEIGHTS_PATH
+from sglang.test.ascend.test_mmlu import TestMMLU
 from sglang.test.ci.ci_register import register_npu_ci
-from sglang.test.few_shot_gsm8k import run_eval as run_gsm8k
-from sglang.test.run_eval import run_eval
-from sglang.test.test_utils import (
-    DEFAULT_URL_FOR_TEST,
-    CustomTestCase,
-    popen_launch_server,
-)
+from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-16-npu-a3", nightly=True)
 
 
-class TestDeepEpDeepseekV32(CustomTestCase):
+class TestDeepEpDeepseekV32(GSM8KAscendMixin, TestMMLU, CustomTestCase):
     """Testcase: Verify that for the DeepSeek V3.2 model in the single-machine colocation scenario,
     its inference accuracy on the MMLU and GSM8K dataset meets the preset standard when the parameter --deepep-mode auto is configured.
 
@@ -24,84 +18,45 @@ class TestDeepEpDeepseekV32(CustomTestCase):
     [Test Target] --moe-a2a-backend deepep;--deepep-mode
     """
 
-    @classmethod
-    def setUpClass(cls):
-        cls.model = DEEPSEEK_V3_2_W8A8_WEIGHTS_PATH
-        cls.base_url = DEFAULT_URL_FOR_TEST
-        cls.process = popen_launch_server(
-            cls.model,
-            cls.base_url,
-            timeout=6000,
-            other_args=[
-                "--trust-remote-code",
-                "--tp-size",
-                "16",
-                "--quantization",
-                "modelslim",
-                "--moe-a2a-backend",
-                "deepep",
-                "--deepep-mode",
-                "auto",
-                "--mem-fraction-static",
-                0.82,
-                "--disable-cuda-graph",
-                "--disable-radix-cache",
-                "--context-length",
-                40960,
-                "--max-prefill-tokens",
-                40960,
-                "--max-total-tokens",
-                40960,
-            ],
-            env={
-                "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
-                "STREAMS_PER_DEVICE": "32",
-                "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "16",
-                "HCCL_BUFFSIZE": "1600",
-                "HCCL_OP_EXPANSION_MODE": "AIV",
-                "SGLANG_NPU_USE_MLAPO": "0",
-                "SGLANG_NPU_USE_MULTI_STREAM": "1",
-                "TASK_QUEUE_ENABLE": "0",
-                **os.environ,
-            },
-        )
+    model = DEEPSEEK_V3_2_W8A8_WEIGHTS_PATH
 
-    @classmethod
-    def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+    timeout_for_server_launch = 60000
+    other_args = [
+        "--trust-remote-code",
+        "--tp-size",
+        "16",
+        "--quantization",
+        "modelslim",
+        "--moe-a2a-backend",
+        "deepep",
+        "--deepep-mode",
+        "auto",
+        "--mem-fraction-static",
+        0.82,
+        "--disable-cuda-graph",
+        "--disable-radix-cache",
+        "--context-length",
+        40960,
+        "--max-prefill-tokens",
+        40960,
+        "--max-total-tokens",
+        40960,
+    ]
 
-    def test_mmlu(self):
-        expect_score = 0.85
-        args = SimpleNamespace(
-            base_url=self.base_url,
-            model=self.model,
-            eval_name="mmlu",
-            num_examples=128,
-            num_threads=32,
-        )
-        print("Starting mmlu test...")
-        metrics = run_eval(args)
-        self.assertGreater(metrics["score"], expect_score)
+    env = {
+        **os.environ,
+        "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
+        "STREAMS_PER_DEVICE": "32",
+        "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "16",
+        "HCCL_BUFFSIZE": "1600",
+        "HCCL_OP_EXPANSION_MODE": "AIV",
+        "SGLANG_NPU_USE_MLAPO": "0",
+        "SGLANG_NPU_USE_MULTI_STREAM": "1",
+        "TASK_QUEUE_ENABLE": "0",
+    }
 
-    def test_gsm8k(self):
-        expect_accuracy = 0.95
-        args = SimpleNamespace(
-            num_shots=8,
-            data_path=None,
-            timeout=60000,
-            num_questions=200,
-            max_new_tokens=512,
-            parallel=128,
-            host="http://127.0.0.1",
-            port=int(self.base_url.split(":")[-1]),
-        )
-        print("Starting gsm8k test...")
-        metrics = run_gsm8k(args)
-        self.assertGreaterEqual(
-            metrics["accuracy"],
-            expect_accuracy,
-            f'Accuracy of {self.model} is {str(metrics["accuracy"])}, is lower than {expect_accuracy}',
-        )
+    accuracy = 0.95  # Test GSM8K accuracy ≥0.95
+    accuracy_mmlu = 0.85  # Test MMLU accuracy ≥0.85
 
 
 if __name__ == "__main__":