ci: migrate MLA tests to test/registered/mla/ (#15798)

2025-12-24 23:29:30 -08:00
parent ccf2330b28
commit 72a980c6d5
8 changed files with 37 additions and 12 deletions
@@ -0,0 +1,61 @@
+import unittest
+from types import SimpleNamespace
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
+from sglang.test.run_eval import run_eval
+from sglang.test.test_utils import (
+    DEFAULT_MLA_MODEL_NAME_FOR_TEST,
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    popen_launch_server,
+)
+
+# MLA attention test with MGSM evaluation
+register_cuda_ci(est_time=194, suite="stage-b-test-small-1-gpu")
+register_amd_ci(
+    est_time=242,
+    suite="stage-a-test-1",
+    disabled="see https://github.com/sgl-project/sglang/issues/13107",
+)
+
+
+class TestMLA(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = DEFAULT_MLA_MODEL_NAME_FOR_TEST
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=[
+                "--trust-remote-code",
+                "--enable-torch-compile",
+                "--torch-compile-max-bs",
+                "4",
+                "--chunked-prefill-size",
+                "256",
+            ],
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_mgsm_en(self):
+        args = SimpleNamespace(
+            base_url=self.base_url,
+            model=self.model,
+            eval_name="mgsm_en",
+            num_examples=None,
+            num_threads=1024,
+        )
+
+        metrics = run_eval(args)
+        self.assertGreater(metrics["score"], 0.8)
+
+
+if __name__ == "__main__":
+    unittest.main()