Add Mimo-v2-flash model to ci test (#15887)

python/sglang/test/kits/gsm8k_accuracy_kit.py

@@ -9,6 +9,8 @@ from sglang.test.few_shot_gsm8k import run_eval as run_eval_gsm8k
 class GSM8KMixin:
     gsm8k_accuracy_thres: float
     gsm8k_accept_length_thres: Optional[float] = None
+    gsm8k_num_questions: int = 200
+    gsm8k_parallel: int = 128
 
     def test_gsm8k(self):
         requests.get(self.base_url + "/flush_cache")
@@ -16,9 +18,9 @@ class GSM8KMixin:
         args = SimpleNamespace(
             num_shots=5,
             data_path=None,
-            num_questions=200,
+            num_questions=self.gsm8k_num_questions,
             max_new_tokens=512,
-            parallel=128,
+            parallel=self.gsm8k_parallel,
             host="http://127.0.0.1",
             port=int(self.base_url.split(":")[-1]),
         )

python/sglang/test/kits/spec_decoding_kit.py

@@ -0,0 +1,23 @@
+from sglang.test.send_one import BenchArgs, send_one_prompt
+from sglang.test.test_utils import is_in_ci, write_github_step_summary
+
+
+class SpecDecodingMixin:
+    bs_1_speed_thres: float
+    accept_length_thres: float
+
+    def test_bs_1_speed(self):
+        args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
+        acc_length, speed = send_one_prompt(args)
+
+        print(f"{acc_length=:.2f} {speed=:.2f}")
+
+        if is_in_ci():
+            write_github_step_summary(
+                f"### test_bs_1_speed ({self.model})\n"
+                f"{acc_length=:.2f}\n"
+                f"{speed=:.2f} token/s\n"
+            )
+
+        self.assertGreater(acc_length, self.accept_length_thres)
+        self.assertGreater(speed, self.bs_1_speed_thres)

test/srt/models/test_mimo_models.py

@@ -0,0 +1,47 @@
+import unittest
+
+from sglang.test.kits.gsm8k_accuracy_kit import GSM8KMixin
+from sglang.test.kits.spec_decoding_kit import SpecDecodingMixin
+from sglang.test.server_fixtures.default_fixture import DefaultServerBase
+
+
+class TestMiMoV2Flash(GSM8KMixin, SpecDecodingMixin, DefaultServerBase):
+    gsm8k_accuracy_thres = 0.75
+    gsm8k_num_questions = 1319
+    gsm8k_parallel = 1319
+    model = "XiaomiMiMo/MiMo-V2-Flash"
+
+    other_args = [
+        "--tp",
+        "4",
+        "--dp",
+        "2",
+        "--enable-dp-attention",
+        "--trust-remote-code",
+        "--attention-backend",
+        "fa3",
+        "--max-running-requests",
+        "128",
+        "--cuda-graph-max-bs",
+        "64",
+        "--mem-fraction-static",
+        "0.75",
+        "--speculative-algorithm",
+        "EAGLE",
+        "--speculative-num-steps",
+        "3",
+        "--speculative-eagle-topk",
+        "1",
+        "--speculative-num-draft-tokens",
+        "4",
+        "--enable-multi-layer-eagle",
+        "--model-loader-extra-config",
+        '{"enable_multithread_load": true,"num_threads": 64}',
+    ]
+
+    bs_1_speed_thres = 170
+    accept_length_thres = 3.2
+
+
+if __name__ == "__main__":
+    unittest.main()

test/srt/run_suite.py

@@ -139,6 +139,7 @@ suites = {
         TestFile("models/test_kimi_k2_models.py", 200),
         TestFile("test_deepseek_v32_basic.py", 275),
         TestFile("test_deepseek_v32_mtp.py", 275),
+        TestFile("models/test_mimo_models.py", 200),
     ],
     "per-commit-8-gpu-h20": [
         TestFile("quant/test_w4a8_deepseek_v3.py", 520),