Update ci test and doc for MTP api change (#5952)

test/srt/test_full_deepseek_v3.py

@@ -80,8 +80,6 @@ class TestDeepseekV3MTP(CustomTestCase):
             "--trust-remote-code",
             "--speculative-algorithm",
             "EAGLE",
-            "--speculative-draft",
-            "lmsys/DeepSeek-V3-0324-NextN",
             "--speculative-num-steps",
             "3",
             "--speculative-eagle-topk",

test/srt/test_mla_deepseek_v3.py

@@ -50,6 +50,63 @@ class TestMLADeepseekV3(CustomTestCase):
 
 
 class TestDeepseekV3MTP(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = "lmsys/sglang-ci-dsv3-test"
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        other_args = [
+            "--trust-remote-code",
+            "--cuda-graph-max-bs",
+            "2",
+            "--disable-radix",
+            "--enable-torch-compile",
+            "--torch-compile-max-bs",
+            "1",
+            "--speculative-algorithm",
+            "EAGLE",
+            "--speculative-num-steps",
+            "2",
+            "--speculative-eagle-topk",
+            "4",
+            "--speculative-num-draft-tokens",
+            "4",
+        ]
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=other_args,
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_gsm8k(self):
+        requests.get(self.base_url + "/flush_cache")
+
+        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_few_shot_gsm8k(args)
+        print(metrics)
+
+        self.assertGreater(metrics["accuracy"], 0.60)
+
+        server_info = requests.get(self.base_url + "/get_server_info")
+        avg_spec_accept_length = server_info.json()["avg_spec_accept_length"]
+        print(f"{avg_spec_accept_length=}")
+        self.assertGreater(avg_spec_accept_length, 2.5)
+
+
+# compatible with old APIs
+class TestDeepseekV3MTPWithDraft(CustomTestCase):
     @classmethod
     def setUpClass(cls):
         cls.model = "lmsys/sglang-ci-dsv3-test"

test/srt/test_mla_flashinfer.py

@@ -118,8 +118,6 @@ class TestFlashinferMLAMTP(CustomTestCase):
                     "1",
                     "--speculative-algorithm",
                     "EAGLE",
-                    "--speculative-draft",
-                    "lmsys/sglang-ci-dsv3-test-NextN",
                     "--speculative-num-steps",
                     "3",
                     "--speculative-eagle-topk",

test/srt/test_mla_int8_deepseek_v3.py

@@ -162,8 +162,6 @@ class TestDeepseekV3MTPBlockInt8(CustomTestCase):
                     "1",
                     "--speculative-algorithm",
                     "EAGLE",
-                    "--speculative-draft",
-                    "sgl-project/sglang-ci-dsv3-block-int8-test-NextN",
                     "--speculative-num-steps",
                     "2",
                     "--speculative-eagle-topk",