[NPU] update nightly tests (#17952)

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: cy <chenyang08056032@163.com>

test/registered/ascend/llm_models/test_ascend_afm_4_5b.py

@@ -1,15 +1,22 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import AFM_4_5B_BASE_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/arcee-ai/AFM-4.5B-Base"
-    accuracy = 0.00
+class TestAFM(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the arcee-ai/AFM-4.5B-Base model on the GSM8K dataset is no less than 0.375.
+
+    [Test Category] Model
+    [Test Target] arcee-ai/AFM-4.5B-Base
+    """
+
+    model = AFM_4_5B_BASE_WEIGHTS_PATH
+    accuracy = 0.375
 
 
 if __name__ == "__main__":

test/registered/ascend/llm_models/test_ascend_baichuan2_13b_chat.py

@@ -1,6 +1,7 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import BAICHUAN2_13B_CHAT_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
@@ -8,7 +9,13 @@ register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
 class TestBaichuan(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/baichuan-inc/Baichuan2-13B-Chat"
+    """Testcase: Verify that the inference accuracy of the baichuan-inc/Baichuan2-13B-Chat model on the GSM8K dataset is no less than 0.48.
+
+    [Test Category] Model
+    [Test Target] baichuan-inc/Baichuan2-13B-Chat
+    """
+
+    model = BAICHUAN2_13B_CHAT_WEIGHTS_PATH
     accuracy = 0.48
     other_args = [
         "--trust-remote-code",

test/registered/ascend/llm_models/test_ascend_charglm2_6b.py

@@ -1,14 +1,21 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import CHATGLM2_6B_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/ZhipuAI/chatglm2-6b"
+class TestChatGlm2(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the ZhipuAI/chatglm2-6b model on the GSM8K dataset is no less than 0.25.
+
+    [Test Category] Model
+    [Test Target] ZhipuAI/chatglm2-6b
+    """
+
+    model = CHATGLM2_6B_WEIGHTS_PATH
     accuracy = 0.25
     other_args = [
         "--trust-remote-code",

test/registered/ascend/llm_models/test_ascend_deepseek_v3_2_exp_w8a8.py

@@ -1,15 +1,28 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import DEEPSEEK_V3_2_EXP_W8A8_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-16-npu-a3", nightly=True)
+register_npu_ci(
+    est_time=400,
+    suite="nightly-16-npu-a3",
+    nightly=True,
+    disabled="run failed",
+)
 
 
-class TestDeepSeekV3_2ExpW8A8(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/DeepSeek-V3.2-Exp-W8A8"
-    accuracy = 0.51
+class TestDeepSeekV32(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the vllm-ascend/DeepSeek-V3.2-Exp-W8A8 model on the GSM8K dataset is no less than 0.5.
+
+    [Test Category] Model
+    [Test Target] vllm-ascend/DeepSeek-V3.2-Exp-W8A8
+    """
+
+    model = DEEPSEEK_V3_2_EXP_W8A8_WEIGHTS_PATH
+    accuracy = 0.5
+    timeout_for_server_launch = 3000
     other_args = [
         "--trust-remote-code",
         "--mem-fraction-static",

test/registered/ascend/llm_models/test_ascend_exaone_3.py

@@ -1,15 +1,22 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import EXAONE_3_5_7_8B_INSTRUCT_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct"
-    accuracy = 0.00
+class TestEXAONE(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct model on the GSM8K dataset is no less than 0.8.
+
+    [Test Category] Model
+    [Test Target] LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct
+    """
+
+    model = EXAONE_3_5_7_8B_INSTRUCT_WEIGHTS_PATH
+    accuracy = 0.8
     other_args = [
         "--trust-remote-code",
         "--mem-fraction-static",

test/registered/ascend/llm_models/test_ascend_glm4_9b_chat.py

@@ -1,15 +1,27 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import GLM_4_9B_CHAT_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+register_npu_ci(
+    est_time=400,
+    suite="nightly-1-npu-a3",
+    nightly=True,
+    disabled="run failed",
+)
 
 
 class TestGLM49BChat(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/ZhipuAI/glm-4-9b-chat"
-    accuracy = 0.00
+    """Testcase: Verify that the inference accuracy of the ZhipuAI/glm-4-9b-chat model on the GSM8K dataset is no less than 0.79.
+
+    [Test Category] Model
+    [Test Target] ZhipuAI/glm-4-9b-chat
+    """
+
+    model = GLM_4_9B_CHAT_WEIGHTS_PATH
+    accuracy = 0.79
 
 
 if __name__ == "__main__":

test/registered/ascend/llm_models/test_ascend_granite_3_0_3b_a800m.py

@@ -1,17 +1,24 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import (
+    GRANITE_3_0_3B_A800M_INSTRUCT_WEIGHTS_PATH,
+)
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = (
-        "/root/.cache/modelscope/hub/models/ibm-granite/granite-3.0-3b-a800m-instruct"
-    )
-    accuracy = 0.00
+class TestGranite(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the ibm-granite/granite-3.0-3b-a800m-instruct model on the GSM8K dataset is no less than 0.38.
+
+    [Test Category] Model
+    [Test Target] ibm-granite/granite-3.0-3b-a800m-instruct
+    """
+
+    model = GRANITE_3_0_3B_A800M_INSTRUCT_WEIGHTS_PATH
+    accuracy = 0.38
 
 
 if __name__ == "__main__":

test/registered/ascend/llm_models/test_ascend_granite_3_1_8b.py

@@ -1,14 +1,21 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import GRANITE_3_1_8B_INSTRUCT_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/ibm-granite/granite-3.1-8b-instruct"
+class TestGranite(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the ibm-granite/granite-3.1-8b-instruct model on the GSM8K dataset is no less than 0.695.
+
+    [Test Category] Model
+    [Test Target] ibm-granite/granite-3.1-8b-instruct
+    """
+
+    model = GRANITE_3_1_8B_INSTRUCT_WEIGHTS_PATH
     accuracy = 0.695
 
 

test/registered/ascend/llm_models/test_ascend_internlm2_7b.py

@@ -1,15 +1,22 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import INTERNLM2_7B_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/Shanghai_AI_Laboratory/internlm2-7b"
-    accuracy = 0.6
+class TestInternlm2(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the Shanghai_AI_Laboratory/internlm2-7b model on the GSM8K dataset is no less than 0.585.
+
+    [Test Category] Model
+    [Test Target] Shanghai_AI_Laboratory/internlm2-7b
+    """
+
+    model = INTERNLM2_7B_WEIGHTS_PATH
+    accuracy = 0.585
 
 
 if __name__ == "__main__":

test/registered/ascend/llm_models/test_ascend_ling_lite.py

@@ -1,15 +1,32 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import LING_LITE_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/inclusionAI/Ling-lite"
+class TestLingLite(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the inclusionAI/Ling-lite model on the GSM8K dataset is no less than 0.75.
+
+    [Test Category] Model
+    [Test Target] inclusionAI/Ling-lite
+    """
+
+    model = LING_LITE_WEIGHTS_PATH
     accuracy = 0.75
+    other_args = [
+        "--trust-remote-code",
+        "--mem-fraction-static",
+        "0.8",
+        "--attention-backend",
+        "ascend",
+        "--disable-cuda-graph",
+        "--tp-size",
+        "2",
+    ]
 
 
 if __name__ == "__main__":

test/registered/ascend/llm_models/test_ascend_llama_2_7b.py

@@ -1,14 +1,21 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import LLAMA_2_7B_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/LLM-Research/Llama-2-7B"
+class TestLlama(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the LLM-Research/Llama-2-7B model on the GSM8K dataset is no less than 0.18.
+
+    [Test Category] Model
+    [Test Target] LLM-Research/Llama-2-7B
+    """
+
+    model = LLAMA_2_7B_WEIGHTS_PATH
     accuracy = 0.18
 
 

test/registered/ascend/llm_models/test_ascend_mimo_7b_rl.py

@@ -1,14 +1,21 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import MIMO_7B_RL_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/XiaomiMiMo/MiMo-7B-RL"
+class TestMiMo7BRL(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the XiaomiMiMo/MiMo-7B-RL model on the GSM8K dataset is no less than 0.75.
+
+    [Test Category] Model
+    [Test Target] XiaomiMiMo/MiMo-7B-RL
+    """
+
+    model = MIMO_7B_RL_WEIGHTS_PATH
     accuracy = 0.75
 
 

test/registered/ascend/llm_models/test_ascend_minicpm3_4b.py

@@ -1,14 +1,26 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import MINICPM3_4B_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
-register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+register_npu_ci(
+    est_time=400,
+    suite="nightly-1-npu-a3",
+    nightly=True,
+    disabled="run failed",
+)
 
 
 class TestMiniCPM3(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/OpenBMB/MiniCPM3-4B"
+    """Testcase: Verify that the inference accuracy of the OpenBMB/MiniCPM3-4B model on the GSM8K dataset is no less than 0.69.
+
+    [Test Category] Model
+    [Test Target] OpenBMB/MiniCPM3-4B
+    """
+
+    model = MINICPM3_4B_WEIGHTS_PATH
     accuracy = 0.69
     other_args = [
         "--trust-remote-code",

test/registered/ascend/llm_models/test_ascend_mistral_7b.py

@@ -1,6 +1,7 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import MISTRAL_7B_INSTRUCT_V0_2_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
@@ -8,7 +9,13 @@ register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
 class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/mistralai/Mistral-7B-Instruct-v0.2"
+    """Testcase: Verify that the inference accuracy of the mistralai/Mistral-7B-Instruct-v0.2 model on the GSM8K dataset is no less than 0.375.
+
+    [Test Category] Model
+    [Test Target] mistralai/Mistral-7B-Instruct-v0.2
+    """
+
+    model = MISTRAL_7B_INSTRUCT_V0_2_WEIGHTS_PATH
     accuracy = 0.375
 
 

test/registered/ascend/llm_models/test_ascend_persimmon_8b_chat.py

@@ -1,14 +1,21 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import PERSIMMON_8B_CHAT_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/Howeee/persimmon-8b-chat"
+class TestPersimmon8BChat(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the Howeee/persimmon-8b-chat model on the GSM8K dataset is no less than 0.17.
+
+    [Test Category] Model
+    [Test Target] Howeee/persimmon-8b-chat
+    """
+
+    model = PERSIMMON_8B_CHAT_WEIGHTS_PATH
     accuracy = 0.17
 
 

test/registered/ascend/llm_models/test_ascend_phi_4_multimodal.py

@@ -1,14 +1,21 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import PHI_4_MULTIMODAL_INSTRUCT_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/LLM-Research/Phi-4-multimodal-instruct"
+class TestPhi4(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the microsoft/Phi-4-multimodal-instruct model on the GSM8K dataset is no less than 0.8.
+
+    [Test Category] Model
+    [Test Target] microsoft/Phi-4-multimodal-instruct
+    """
+
+    model = PHI_4_MULTIMODAL_INSTRUCT_WEIGHTS_PATH
     accuracy = 0.8
 
 

test/registered/ascend/llm_models/test_ascend_qwen3_0_6b.py

@@ -0,0 +1,30 @@
+import unittest
+
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import QWEN3_0_6B_WEIGHTS_PATH
+from sglang.test.ci.ci_register import register_npu_ci
+from sglang.test.test_utils import CustomTestCase
+
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
+
+class TestQwen306B(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the Qwen/Qwen3-0.6B model on the GSM8K dataset is no less than 0.38.
+
+    [Test Category] Model
+    [Test Target] Qwen/Qwen3-0.6B
+    """
+
+    model = QWEN3_0_6B_WEIGHTS_PATH
+    accuracy = 0.38
+    other_args = [
+        "--chunked-prefill-size",
+        256,
+        "--attention-backend",
+        "ascend",
+        "--disable-cuda-graph",
+    ]
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_qwen3_235b_a22b_w8a8.py

@@ -0,0 +1,35 @@
+import unittest
+
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import QWEN3_235B_A22B_W8A8_WEIGHTS_PATH
+from sglang.test.ci.ci_register import register_npu_ci
+from sglang.test.test_utils import CustomTestCase
+
+register_npu_ci(est_time=400, suite="nightly-8-npu-a3", nightly=True)
+
+
+class TestQwen3235BA22BW8A8(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the vllm-ascend/Qwen3-235B-A22B-W8A8 model on the GSM8K dataset is no less than 0.955.
+
+    [Test Category] Model
+    [Test Target] vllm-ascend/Qwen3-235B-A22B-W8A8
+    """
+
+    model = QWEN3_235B_A22B_W8A8_WEIGHTS_PATH
+    accuracy = 0.955
+    other_args = [
+        "--trust-remote-code",
+        "--mem-fraction-static",
+        "0.8",
+        "--attention-backend",
+        "ascend",
+        "--disable-cuda-graph",
+        "--tp-size",
+        "8",
+        "--quantization",
+        "modelslim",
+    ]
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_qwen3_30b.py

@@ -0,0 +1,39 @@
+import unittest
+
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import (
+    Qwen3_30B_A3B_Instruct_2507_WEIGHTS_PATH,
+)
+from sglang.test.ci.ci_register import register_npu_ci
+from sglang.test.test_utils import CustomTestCase
+
+register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
+
+
+class TestQwen330B(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the Qwen/Qwen3-30B-A3B-Instruct-2507 model on the GSM8K dataset is no less than 0.90.
+
+    [Test Category] Model
+    [Test Target] Qwen/Qwen3-30B-A3B-Instruct-2507
+    """
+
+    model = Qwen3_30B_A3B_Instruct_2507_WEIGHTS_PATH
+    accuracy = 0.90
+    other_args = [
+        "--trust-remote-code",
+        "--mem-fraction-static",
+        0.7,
+        "--max-running-requests",
+        32,
+        "--attention-backend",
+        "ascend",
+        "--disable-cuda-graph",
+        "--cuda-graph-max-bs",
+        32,
+        "--tp-size",
+        2,
+    ]
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_qwen3_32b.py

@@ -0,0 +1,38 @@
+import unittest
+
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import QWEN3_32B_WEIGHTS_PATH
+from sglang.test.ci.ci_register import register_npu_ci
+from sglang.test.test_utils import CustomTestCase
+
+register_npu_ci(
+    est_time=400,
+    suite="nightly-4-npu-a3",
+    nightly=True,
+    disabled="run failed",
+)
+
+
+class TestQwen332B(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the Qwen/Qwen3-32B model on the GSM8K dataset is no less than 0.88.
+
+    [Test Category] Model
+    [Test Target] Qwen/Qwen3-32B
+    """
+
+    model = QWEN3_32B_WEIGHTS_PATH
+    accuracy = 0.88
+    other_args = [
+        "--trust-remote-code",
+        "--mem-fraction-static",
+        "0.8",
+        "--attention-backend",
+        "ascend",
+        "--disable-cuda-graph",
+        "--tp-size",
+        "4",
+    ]
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_qwen3_coder_480b_a35b.py

@@ -0,0 +1,43 @@
+import unittest
+
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import (
+    QWEN3_CODER_480B_A35B_INSTRUCT_W8A8_QUAROT_WEIGHTS_PATH,
+)
+from sglang.test.ci.ci_register import register_npu_ci
+from sglang.test.test_utils import CustomTestCase
+
+register_npu_ci(
+    est_time=400,
+    suite="nightly-16-npu-a3",
+    nightly=True,
+    disabled="run failed",
+)
+
+
+class TestQwen3Coder480BA35B(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the Qwen3-Coder-480B-A35B-Instruct-w8a8-QuaRot model on the GSM8K dataset is no less than 0.94.
+
+    [Test Category] Model
+    [Test Target] Qwen3-Coder-480B-A35B-Instruct-w8a8-QuaRot
+    """
+
+    model = QWEN3_CODER_480B_A35B_INSTRUCT_W8A8_QUAROT_WEIGHTS_PATH
+    accuracy = 0.94
+    timeout_for_server_launch = 3000
+    other_args = [
+        "--trust-remote-code",
+        "--mem-fraction-static",
+        "0.8",
+        "--attention-backend",
+        "ascend",
+        "--disable-cuda-graph",
+        "--tp-size",
+        "16",
+        "--quantization",
+        "modelslim",
+    ]
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_qwq_32b_w8a8.py

@@ -0,0 +1,35 @@
+import unittest
+
+from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import QWQ_32B_W8A8_WEIGHTS_PATH
+from sglang.test.ci.ci_register import register_npu_ci
+from sglang.test.test_utils import CustomTestCase
+
+register_npu_ci(est_time=400, suite="nightly-2-npu-a3", nightly=True)
+
+
+class TestQWQ32BW8A8(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the vllm-ascend/QWQ-32B-W8A8 model on the GSM8K dataset is no less than 0.59.
+
+    [Test Category] Model
+    [Test Target] vllm-ascend/QWQ-32B-W8A8
+    """
+
+    model = QWQ_32B_W8A8_WEIGHTS_PATH
+    accuracy = 0.59
+    other_args = [
+        "--trust-remote-code",
+        "--mem-fraction-static",
+        "0.8",
+        "--attention-backend",
+        "ascend",
+        "--disable-cuda-graph",
+        "--tp-size",
+        "2",
+        "--quantization",
+        "modelslim",
+    ]
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_smollm_1_7b.py

@@ -1,14 +1,21 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import SMOLLM_1_7B_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
 register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
-class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/HuggingFaceTB/SmolLM-1.7B"
+class TestSmolLM(GSM8KAscendMixin, CustomTestCase):
+    """Testcase: Verify that the inference accuracy of the HuggingFaceTB/SmolLM-1.7B model on the GSM8K dataset is no less than 0.05.
+
+    [Test Category] Model
+    [Test Target] HuggingFaceTB/SmolLM-1.7B
+    """
+
+    model = SMOLLM_1_7B_WEIGHTS_PATH
     accuracy = 0.05
     other_args = [
         "--trust-remote-code",

test/registered/ascend/llm_models/test_ascend_stablelm-2-1_6b.py

@@ -1,6 +1,7 @@
 import unittest
 
 from sglang.test.ascend.gsm8k_ascend_mixin import GSM8KAscendMixin
+from sglang.test.ascend.test_ascend_utils import STABLELM_2_1_6B_WEIGHTS_PATH
 from sglang.test.ci.ci_register import register_npu_ci
 from sglang.test.test_utils import CustomTestCase
 
@@ -8,7 +9,13 @@ register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
 
 
 class TestStablelm(GSM8KAscendMixin, CustomTestCase):
-    model = "/root/.cache/modelscope/hub/models/stabilityai/stablelm-2-1_6b"
+    """Testcase: Verify that the inference accuracy of the stabilityai/stablelm-2-1_6b model on the GSM8K dataset is no less than 0.195.
+
+    [Test Category] Model
+    [Test Target] stabilityai/stablelm-2-1_6b
+    """
+
+    model = STABLELM_2_1_6B_WEIGHTS_PATH
     accuracy = 0.195
     other_args = [
         "--trust-remote-code",