[Qwen3.5] Support Qwen3.5 Pipeline Parallelism (#19670)

Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>

test/registered/distributed/test_pp_single_node.py

@@ -350,6 +350,60 @@ class TestQwenMoePPAccuracy(unittest.TestCase):
         )
 
 
+class TestQwen35PPAccuracy(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.base_url = "http://127.0.0.1:23337"  # different ports to avoid conflicts
+        cls.model_name = (
+            "Qwen/Qwen3.5-35B-A3B"  # replace with your Qwen Model if needed
+        )
+
+    def run_gsm8k_test(self, pp_size):
+        process = popen_launch_server(
+            self.model_name,
+            self.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=[
+                "--pp-size",
+                pp_size,
+                "--chunked-prefill-size",
+                256,
+            ],
+        )
+
+        try:
+            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)
+            time.sleep(5)
+            return metrics
+        finally:
+            kill_process_tree(process.pid)
+
+    def test_pp_consistency(self):
+        baseline = self.run_gsm8k_test(pp_size=1)
+        pp_metrics = self.run_gsm8k_test(pp_size=2)
+
+        print(f"[Qwen35 PP Comparison] Baseline: {baseline} | PP: {pp_metrics}")
+
+        self.assertGreaterEqual(baseline["accuracy"], 0.83)
+        self.assertGreaterEqual(
+            pp_metrics["accuracy"],
+            baseline["accuracy"] - 0.02,
+            msg=(
+                f"PP accuracy dropped more than 1% compared to baseline. "
+                f"Baseline: {baseline['accuracy']:.2%}, PP: {pp_metrics['accuracy']:.2%}"
+            ),
+        )
+
+
 class TestFixedBugs(unittest.TestCase):
     def test_chunked_prefill_with_small_bs(self):
         model = DEFAULT_MODEL_NAME_FOR_TEST