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

test/registered/ascend/embedding_models/test_ascend_embedding_models.py

@@ -0,0 +1,108 @@
+import multiprocessing as mp
+import unittest
+from typing import Optional
+
+import torch
+from transformers import AutoConfig, AutoTokenizer
+
+from sglang.test.ci.ci_register import register_npu_ci
+from sglang.test.runners import DEFAULT_PROMPTS, HFRunner, SRTRunner
+from sglang.test.test_utils import CustomTestCase, get_similarities
+
+register_npu_ci(
+    est_time=400,
+    suite="nightly-1-npu-a3",
+    nightly=True,
+    disabled="embeddings are not all close",
+)
+
+
+MODELS = [
+    ("/root/.cache/modelscope/hub/models/iic/gte_Qwen2-1.5B-instruct", 1, 1e-5),
+    ("/root/.cache/modelscope/hub/models/Qwen/Qwen3-Embedding-8B", 1, 1e-5),
+]
+TORCH_DTYPES = [torch.bfloat16]
+
+
+class TestEmbeddingModels(CustomTestCase):
+
+    @classmethod
+    def setUpClass(cls):
+        mp.set_start_method("spawn", force=True)
+
+    def _truncate_prompts(self, prompts, model_path):
+        config = AutoConfig.from_pretrained(model_path)
+        max_length = getattr(config, "max_position_embeddings", 2048)
+
+        tokenizer = AutoTokenizer.from_pretrained(model_path)
+
+        truncated_prompts = []
+        for prompt in prompts:
+            tokens = tokenizer(prompt, return_tensors="pt", truncation=False)
+            if len(tokens.input_ids[0]) > max_length:
+                truncated_text = tokenizer.decode(
+                    tokens.input_ids[0][: max_length - 1], skip_special_tokens=True
+                )
+                truncated_prompts.append(truncated_text)
+            else:
+                truncated_prompts.append(prompt)
+        return truncated_prompts
+
+    def assert_close_prefill_logits(
+        self,
+        prompts,
+        model_path,
+        tp_size,
+        torch_dtype,
+        prefill_tolerance,
+        matryoshka_dim: Optional[int] = None,
+    ) -> None:
+        truncated_prompts = self._truncate_prompts(prompts, model_path)
+
+        with HFRunner(
+            model_path,
+            torch_dtype=torch_dtype,
+            model_type="embedding",
+            matryoshka_dim=matryoshka_dim,
+        ) as hf_runner:
+            hf_outputs = hf_runner.forward(truncated_prompts)
+
+        attention_backend = "ascend"
+        with SRTRunner(
+            model_path,
+            tp_size=tp_size,
+            torch_dtype=torch_dtype,
+            model_type="embedding",
+            attention_backend=attention_backend,
+            json_model_override_args=(
+                {"matryoshka_dimensions": [matryoshka_dim]} if matryoshka_dim else None
+            ),
+        ) as srt_runner:
+            srt_outputs = srt_runner.forward(
+                truncated_prompts, dimensions=matryoshka_dim
+            )
+
+        for i in range(len(prompts)):
+            hf_logits = torch.Tensor(hf_outputs.embed_logits[i])
+            srt_logits = torch.Tensor(srt_outputs.embed_logits[i])
+
+            similarity = torch.tensor(get_similarities(hf_logits, srt_logits))
+            print("similarity diff", abs(similarity - 1))
+
+            if len(prompts[i]) <= 1000:
+                assert torch.all(
+                    abs(similarity - 1) < prefill_tolerance
+                ), "embeddings are not all close"
+
+    def test_prefill_logits(self):
+        models_to_test = MODELS
+
+        for model, tp_size, prefill_tolerance in models_to_test:
+            for torch_dtype in TORCH_DTYPES:
+                self.assert_close_prefill_logits(
+                    DEFAULT_PROMPTS, model, tp_size, torch_dtype, prefill_tolerance
+                )
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_afm_4_5b.py

@@ -0,0 +1,16 @@
+import unittest
+
+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
+
+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
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_baichuan2_13b_chat.py

@@ -0,0 +1,21 @@
+import unittest
+
+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
+
+register_npu_ci(
+    est_time=400,
+    suite="nightly-1-npu-a3",
+    nightly=True,
+    disabled="The accuracy test result is 0.",
+)
+
+
+class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
+    model = "/root/.cache/modelscope/hub/models/baichuan-inc/Baichuan2-13B-Chat"
+    accuracy = 0.00
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_c4ai_command_r_v01.py

@@ -0,0 +1,91 @@
+import os
+import unittest
+from types import SimpleNamespace
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_npu_ci
+from sglang.test.few_shot_gsm8k import run_eval
+from sglang.test.test_utils import (
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    popen_launch_server,
+)
+
+register_npu_ci(
+    est_time=400,
+    suite="nightly-2-npu-a3",
+    nightly=True,
+    disabled="The accuracy test result is 0.",
+)
+
+
+class TestC4AI(CustomTestCase):
+    model = "/root/.cache/modelscope/hub/models/CohereForAI/c4ai-command-r-v01"
+    accuracy = 0.05
+
+    @classmethod
+    def setUpClass(cls):
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        chat_template_path = "/__w/sglang/sglang/test/nightly/ascend/llm_models/tool_chat_template_c4ai_command_r_v01.jinja"
+
+        other_args = [
+            "--trust-remote-code",
+            "--mem-fraction-static",
+            "0.8",
+            "--attention-backend",
+            "ascend",
+            "--disable-cuda-graph",
+            "--chat-template",
+            chat_template_path,
+            "--tp-size",
+            "2",
+            "--dtype",
+            "bfloat16",
+        ]
+        env = os.environ.copy()
+        env.update(
+            {
+                "PYTORCH_NPU_ALLOC_CONF": "expandable_segments:True",
+                "ASCEND_MF_STORE_URL": "tcp://127.0.0.1:24666",
+                "HCCL_BUFFSIZE": "200",
+                "SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "24",
+                "USE_VLLM_CUSTOM_ALLREDUCE": "1",
+                "HCCL_EXEC_TIMEOUT": "200",
+                "STREAMS_PER_DEVICE": "32",
+                "SGLANG_ENABLE_TORCH_COMPILE": "1",
+            }
+        )
+
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=other_args,
+            env=env,
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_gsm8k(self):
+        args = SimpleNamespace(
+            num_shots=5,
+            data_path=None,
+            num_questions=200,
+            max_new_tokens=512,
+            parallel=128,
+            host="http://127.0.0.1",
+            port=int(self.base_url.split(":")[-1]),
+        )
+        metrics = run_eval(args)
+        self.assertGreater(
+            metrics["accuracy"],
+            self.accuracy,
+            f'Accuracy of {self.model} is {str(metrics["accuracy"])}, is lower than {self.accuracy}',
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_charglm2_6b.py

@@ -0,0 +1,26 @@
+import unittest
+
+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
+
+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"
+    accuracy = 0.25
+    other_args = [
+        "--trust-remote-code",
+        "--mem-fraction-static",
+        "0.8",
+        "--attention-backend",
+        "ascend",
+        "--disable-cuda-graph",
+        "--dtype",
+        "bfloat16",
+    ]
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_exaone_3.py

@@ -0,0 +1,26 @@
+import unittest
+
+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
+
+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
+    other_args = [
+        "--trust-remote-code",
+        "--mem-fraction-static",
+        "0.8",
+        "--attention-backend",
+        "ascend",
+        "--disable-cuda-graph",
+        "--dtype",
+        "bfloat16",
+    ]
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_gemma_3_1b_it.py

@@ -0,0 +1,21 @@
+import unittest
+
+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
+
+register_npu_ci(
+    est_time=400,
+    suite="nightly-1-npu-a3",
+    nightly=True,
+    disabled="The accuracy test result is 0.",
+)
+
+
+class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
+    model = "/root/.cache/modelscope/hub/models/LLM-Research/gemma-3-1b-it"
+    accuracy = 0.00
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_glm4_9b_chat.py

@@ -0,0 +1,16 @@
+import unittest
+
+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
+
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
+
+class TestGLM49BChat(GSM8KAscendMixin, CustomTestCase):
+    model = "/root/.cache/modelscope/hub/models/ZhipuAI/glm-4-9b-chat"
+    accuracy = 0.00
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_granite_3_0_3b_a800m.py

@@ -0,0 +1,18 @@
+import unittest
+
+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
+
+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
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_granite_3_1_8b.py

@@ -0,0 +1,16 @@
+import unittest
+
+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
+
+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"
+    accuracy = 0.695
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_internlm2_7b.py

@@ -0,0 +1,16 @@
+import unittest
+
+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
+
+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
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_ling_lite.py

@@ -0,0 +1,16 @@
+import unittest
+
+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
+
+register_npu_ci(est_time=400, suite="nightly-1-npu-a3", nightly=True)
+
+
+class TestMistral7B(GSM8KAscendMixin, CustomTestCase):
+    model = "/root/.cache/modelscope/hub/models/inclusionAI/Ling-lite"
+    accuracy = 0.75
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_llama_2_7b.py

@@ -0,0 +1,16 @@
+import unittest
+
+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
+
+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"
+    accuracy = 0.18
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_mimo_7b_rl.py

@@ -0,0 +1,16 @@
+import unittest
+
+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
+
+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"
+    accuracy = 0.75
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_mistral_7b.py

@@ -0,0 +1,16 @@
+import unittest
+
+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
+
+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"
+    accuracy = 0.375
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_persimmon_8b_chat.py

@@ -0,0 +1,16 @@
+import unittest
+
+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
+
+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"
+    accuracy = 0.17
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_phi_4_multimodal.py

@@ -0,0 +1,16 @@
+import unittest
+
+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
+
+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"
+    accuracy = 0.8
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/test_ascend_smollm_1_7b.py

@@ -0,0 +1,26 @@
+import unittest
+
+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
+
+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"
+    accuracy = 0.05
+    other_args = [
+        "--trust-remote-code",
+        "--mem-fraction-static",
+        "0.8",
+        "--attention-backend",
+        "ascend",
+        "--disable-cuda-graph",
+        "--dtype",
+        "bfloat16",
+    ]
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/llm_models/tool_chat_template_c4ai_command_r_v01.jinja

@@ -0,0 +1 @@
+{{ bos_token }}{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% elif false == true %}{% set loop_messages = messages %}{% set system_message = 'You are Command-R, a brilliant, sophisticated, AI-assistant trained to assist human users by providing thorough responses. You are trained by Cohere.' %}{% else %}{% set loop_messages = messages %}{% set system_message = false %}{% endif %}{% if system_message != false %}{{ '<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>' + system_message + '<|END_OF_TURN_TOKEN|>' }}{% endif %}{% for message in loop_messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ '<|START_OF_TURN_TOKEN|><|USER_TOKEN|>' + content.strip() + '<|END_OF_TURN_TOKEN|>' }}{% elif message['role'] == 'assistant' %}{{ '<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>'  + content.strip() + '<|END_OF_TURN_TOKEN|>' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>' }}{% endif %}

test/registered/ascend/rerank_models/test_ascend_cross_encoder_models.py

@@ -0,0 +1,92 @@
+import multiprocessing as mp
+import unittest
+
+import torch
+
+from sglang.test.ci.ci_register import register_npu_ci
+from sglang.test.runners import TEST_RERANK_QUERY_DOCS, HFRunner, SRTRunner
+from sglang.test.test_utils import CustomTestCase
+
+register_npu_ci(
+    est_time=400,
+    suite="nightly-1-npu-a3",
+    nightly=True,
+    disabled="cross encoder scores are not all close",
+)
+
+MODELS = [
+    ("/root/.cache/modelscope/hub/models/BAAI/bge-reranker-v2-m3", 1, 1e-2),
+]
+ATTENTION_BACKEND = ["ascend"]
+TORCH_DTYPES = [torch.bfloat16]
+
+
+class TestCrossEncoderModels(CustomTestCase):
+
+    @classmethod
+    def setUpClass(cls):
+        mp.set_start_method("spawn", force=True)
+
+    def assert_close_prefill_logits(
+        self,
+        prompts,
+        model_path,
+        tp_size,
+        torch_dtype,
+        score_tolerance,
+        attention_backend,
+    ) -> None:
+        with HFRunner(
+            model_path,
+            torch_dtype=torch_dtype,
+            model_type="cross_encoder",
+        ) as hf_runner:
+            hf_scores = hf_runner.forward(prompts).scores
+
+        with SRTRunner(
+            model_path,
+            tp_size=tp_size,
+            torch_dtype=torch_dtype,
+            model_type="cross_encoder",
+            attention_backend=attention_backend,
+            chunked_prefill_size=-1,
+            disable_radix_cache=True,
+        ) as srt_runner:
+            srt_scores = srt_runner.forward(prompts).scores
+
+        for i in range(len(srt_scores)):
+            score_difference = abs(hf_scores[i] - srt_scores[i])
+
+            assert (
+                score_difference < score_tolerance
+            ), "cross encoder scores are not all close"
+
+    def preprocess_prompts(self, prompt):
+        processed_prompts = []
+        query = prompt["query"]
+        documents = prompt["documents"]
+        for document in documents:
+            processed_prompts.append([query, document])
+
+        return processed_prompts
+
+    def test_prefill_logits(self):
+        models_to_test = MODELS
+
+        for model, tp_size, prefill_tolerance in models_to_test:
+            for attention_backend in ATTENTION_BACKEND:
+                for queryDocs in TEST_RERANK_QUERY_DOCS:
+                    prompts = self.preprocess_prompts(queryDocs)
+                    for torch_dtype in TORCH_DTYPES:
+                        self.assert_close_prefill_logits(
+                            prompts,
+                            model,
+                            tp_size,
+                            torch_dtype,
+                            prefill_tolerance,
+                            attention_backend,
+                        )
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/vlm_models/mmmu-val.yaml

@@ -0,0 +1 @@
+dataset_path: /root/.cache/huggingface/hub/datasets--lmms-lab--MMMU/snapshots/364f2e2eb107b36e07ff4c5a15f5947a759cef47

test/registered/ascend/vlm_models/test_ascend_gemma_3_4b_it.py

@@ -0,0 +1,18 @@
+import unittest
+
+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)
+
+
+class TestGemmaModels(TestVLMModels):
+    model = "/root/.cache/modelscope/hub/models/google/gemma-3-4b-it"
+    mmmu_accuracy = 0.2
+
+    def test_vlm_mmmu_benchmark(self):
+        self._run_vlm_mmmu_test()
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/vlm_models/test_ascend_janus_pro_1b.py

@@ -0,0 +1,18 @@
+import unittest
+
+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)
+
+
+class TestGemmaModels(TestVLMModels):
+    model = "/root/.cache/modelscope/hub/models/deepseek-ai/Janus-Pro-1B"
+    mmmu_accuracy = 0.2
+
+    def test_vlm_mmmu_benchmark(self):
+        self._run_vlm_mmmu_test()
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/vlm_models/test_ascend_janus_pro_7b.py

@@ -0,0 +1,18 @@
+import unittest
+
+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)
+
+
+class TestJanusPro7B(TestVLMModels):
+    model = "/root/.cache/modelscope/hub/models/deepseek-ai/Janus-Pro-7B"
+    mmmu_accuracy = 0.2
+
+    def test_vlm_mmmu_benchmark(self):
+        self._run_vlm_mmmu_test()
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/vlm_models/test_ascend_mimo_vl_7b_rl.py

@@ -0,0 +1,18 @@
+import unittest
+
+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)
+
+
+class TestGemmaModels(TestVLMModels):
+    model = "/root/.cache/modelscope/hub/models/XiaomiMiMo/MiMo-VL-7B-RL"
+    mmmu_accuracy = 0.2
+
+    def test_vlm_mmmu_benchmark(self):
+        self._run_vlm_mmmu_test()
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/vlm_models/test_ascend_minicpm_o_2_6.py

@@ -0,0 +1,18 @@
+import unittest
+
+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)
+
+
+class TestGemmaModels(TestVLMModels):
+    model = "/root/.cache/modelscope/hub/models/openbmb/MiniCPM-o-2_6"
+    mmmu_accuracy = 0.2
+
+    def test_vlm_mmmu_benchmark(self):
+        self._run_vlm_mmmu_test()
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/vlm_models/test_ascend_minicpm_v_2_6.py

@@ -0,0 +1,18 @@
+import unittest
+
+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)
+
+
+class TestGemmaModels(TestVLMModels):
+    model = "/root/.cache/modelscope/hub/models/openbmb/MiniCPM-V-2_6"
+    mmmu_accuracy = 0.2
+
+    def test_vlm_mmmu_benchmark(self):
+        self._run_vlm_mmmu_test()
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/vlm_models/test_ascend_phi4_multimodal_instruct.py

@@ -0,0 +1,18 @@
+import unittest
+
+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)
+
+
+class TestGemmaModels(TestVLMModels):
+    model = "/root/.cache/modelscope/hub/models/microsoft/Phi-4-multimodal-instruct"
+    mmmu_accuracy = 0.2
+
+    def test_vlm_mmmu_benchmark(self):
+        self._run_vlm_mmmu_test()
+
+
+if __name__ == "__main__":
+    unittest.main()

test/registered/ascend/vlm_models/test_ascend_qwen2_5_vl_3b_instruct.py

@@ -0,0 +1,18 @@
+import unittest
+
+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)
+
+
+class TestGemmaModels(TestVLMModels):
+    model = "/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-VL-3B-Instruct"
+    mmmu_accuracy = 0.2
+
+    def test_vlm_mmmu_benchmark(self):
+        self._run_vlm_mmmu_test()
+
+
+if __name__ == "__main__":
+    unittest.main()