Support Eagle2 for Triton backend (#3466)

test/srt/test_eagle_infer.py

@@ -193,5 +193,34 @@ class TestEAGLEServer(unittest.TestCase):
         self.assertGreater(metrics["accuracy"], 0.20)
 
 
+class TestEAGLEServerTriton(TestEAGLEServer):
+    @classmethod
+    def setUpClass(cls):
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        cls.process = popen_launch_server(
+            DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=[
+                "--speculative-algorithm",
+                "EAGLE",
+                "--speculative-draft-model-path",
+                DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST,
+                "--speculative-num-steps",
+                "5",
+                "--speculative-eagle-topk",
+                "8",
+                "--speculative-num-draft-tokens",
+                "64",
+                "--mem-fraction-static",
+                "0.7",
+                "--attention-backend",
+                "triton",
+                # TODO: Support cuda graph
+                "--disable-cuda-graph",
+            ],
+        )
+
+
 if __name__ == "__main__":
     unittest.main()