[PP] Add dynamic chunking PP test (#15395)

Signed-off-by: Shangming Cai <csmthu@gmail.com>

test/srt/run_suite.py

@@ -161,7 +161,7 @@ suites = {
     "per-commit-8-gpu-h20": [
         TestFile("quant/test_w4a8_deepseek_v3.py", 520),
         TestFile("test_disaggregation_different_tp.py", 600),
-        TestFile("test_disaggregation_pp.py", 140),
+        TestFile("test_disaggregation_pp.py", 180),
         TestFile("test_disaggregation_dp_attention.py", 155),
     ],
     "per-commit-4-gpu-b200-stage-b": [

test/srt/test_disaggregation_pp.py

@@ -87,6 +87,80 @@ class TestDisaggregationPrefillPPAccuracy(PDDisaggregationServerBase):
         time.sleep(5)
 
 
+class TestDisaggregationPrefillPPDynamicChunkAccuracy(PDDisaggregationServerBase):
+    @classmethod
+    def setUpClass(cls):
+        super().setUpClass()
+        cls.model = try_cached_model(DEFAULT_MODEL_NAME_FOR_TEST)
+
+        # Non blocking start servers
+        cls.start_prefill()
+        cls.start_decode()
+
+        # Block until both
+        cls.wait_server_ready(cls.prefill_url + "/health")
+        cls.wait_server_ready(cls.decode_url + "/health")
+
+        cls.launch_lb()
+
+    @classmethod
+    def start_prefill(cls):
+        prefill_args = [
+            "--trust-remote-code",
+            "--disaggregation-mode",
+            "prefill",
+            "--tp-size",
+            "2",
+            "--pp-size",
+            "2",
+            "--disable-overlap-schedule",
+            "--enable-dynamic-chunking",
+        ]
+        prefill_args += cls.transfer_backend + cls.rdma_devices
+        cls.process_prefill = popen_launch_pd_server(
+            cls.model,
+            cls.prefill_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=prefill_args,
+        )
+
+    @classmethod
+    def start_decode(cls):
+        decode_args = [
+            "--trust-remote-code",
+            "--disaggregation-mode",
+            "decode",
+            "--tp-size",
+            "2",
+            "--base-gpu-id",
+            "4",
+        ]
+        decode_args += cls.transfer_backend + cls.rdma_devices
+        cls.process_decode = popen_launch_pd_server(
+            cls.model,
+            cls.decode_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=decode_args,
+        )
+
+    def test_gsm8k(self):
+        args = SimpleNamespace(
+            num_shots=5,
+            data_path=None,
+            num_questions=200,
+            max_new_tokens=512,
+            parallel=128,
+            host=f"http://{self.base_host}",
+            port=int(self.lb_port),
+        )
+        metrics = run_eval(args)
+        print(f"{metrics=}")
+
+        self.assertGreater(metrics["accuracy"], 0.24)
+        # Wait a little bit so that the memory check happens.
+        time.sleep(5)
+
+
 class TestDisaggregationDecodePPAccuracy(PDDisaggregationServerBase):
     @classmethod
     def setUpClass(cls):