Re-introduce the unit test of test_mooncake_ep_small (#16019)

test/srt/ep/test_mooncake_ep_small.py

@@ -35,7 +35,7 @@ class TestTP(CustomTestCase):
                 "--mooncake-ib-device",
                 ib_devices,
                 "--moe-a2a-backend",
-                "deepep",
+                "mooncake",
                 "--deepep-mode",
                 "low_latency",
                 "--chunked-prefill-size",
@@ -70,6 +70,7 @@ class TestTP(CustomTestCase):
         self.assertGreater(metrics["accuracy"], 0.60)
 
 
+@unittest.skip("covered in TestMooncakeWithEPLB")
 class TestPureDP(TestTP):
     extra_args = [
         "--tp",
@@ -90,6 +91,7 @@ class TestHybridDPTP(TestTP):
     ]
 
 
+@unittest.skip("covered in TestMooncakeWithEPLB")
 class TestNoGatherdBuffer(TestTP):
     extra_args = [
         "--tp",
@@ -102,6 +104,7 @@ class TestNoGatherdBuffer(TestTP):
     ]
 
 
+@unittest.skip("covered in TestMooncakeWithEPLB")
 class TestTBO(TestTP):
     extra_args = [
         "--tp",
@@ -115,7 +118,7 @@ class TestTBO(TestTP):
     ]
 
 
-class TestMooncakeWitchEPLB(TestTP):
+class TestMooncakeWithEPLB(TestTP):
     extra_args = [
         "--tp",
         "4",

test/srt/run_suite.py

@@ -45,8 +45,7 @@ suites = {
     ],
     "per-commit-4-gpu-deepep": [
         TestFile("ep/test_deepep_small.py", 531),
-        # TODO: Add it back after mooncake supports torch 2.9
-        # TestFile("ep/test_mooncake_ep_small.py", 450),
+        TestFile("ep/test_mooncake_ep_small.py", 660),
     ],
     "per-commit-8-gpu-h200-deepep": [
         TestFile("ep/test_deepep_large.py", 563),