[DeepSeek v3.2] opt Context Parallelism: support fused moe, multi batch and fp8 kvcache (#13959)

test/manual/test_deepseek_v32_cp_single_node.py

@@ -95,5 +95,79 @@ class TestDeepseekV32CP(CustomTestCase):
             self.assertGreater(avg_spec_accept_length, 2.7)
 
 
+class TestDeepseekV32CPMode1(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = FULL_DEEPSEEK_V32_MODEL_PATH
+        cls.base_url = DEFAULT_URL_FOR_TEST
+        other_args = [
+            "--trust-remote-code",
+            "--tp",
+            "8",
+            "--dp",
+            "1",
+            "--enable-dp-attention",
+            "--speculative-algorithm",
+            "EAGLE",
+            "--speculative-num-steps",
+            "3",
+            "--speculative-eagle-topk",
+            "1",
+            "--speculative-num-draft-tokens",
+            "4",
+            "--mem-frac",
+            "0.7",
+            "--cuda-graph-max-bs",
+            "32",
+            "--max-running-requests",
+            "32",
+            "--enable-nsa-prefill-context-parallel",
+            "--nsa-prefill-cp-mode",
+            "round-robin-split",
+        ]
+        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_a_gsm8k(
+        self,
+    ):  # Append an "a" to make this test run first (alphabetically) to warm up the server
+        requests.get(self.base_url + "/flush_cache")
+
+        args = SimpleNamespace(
+            num_shots=8,
+            data_path=None,
+            num_questions=200,
+            parallel=32,
+            max_new_tokens=512,
+            host="http://127.0.0.1",
+            port=int(self.base_url.split(":")[-1]),
+        )
+        metrics = run_eval_few_shot_gsm8k(args)
+        print(f"{metrics=}")
+
+        server_info = requests.get(self.base_url + "/get_server_info")
+        avg_spec_accept_length = server_info.json()["internal_states"][0][
+            "avg_spec_accept_length"
+        ]
+        print(f"{avg_spec_accept_length=}")
+
+        if is_in_ci():
+            write_github_step_summary(
+                f"### test_gsm8k (deepseek-v32 nsa-cp)\n"
+                f'{metrics["accuracy"]=:.3f}\n'
+                f"{avg_spec_accept_length=:.2f}\n"
+            )
+            self.assertGreater(metrics["accuracy"], 0.935)
+            self.assertGreater(avg_spec_accept_length, 2.7)
+
+
 if __name__ == "__main__":
     unittest.main()

test/srt/test_prefill_adder.py

@@ -12,7 +12,7 @@ class TestPrefillAdder(CustomTestCase):
         self.mock_tree_cache = self.create_tree_cache()
         self.mock_token_allocator = self.create_token_allocator()
         patcher = patch(
-            "sglang.srt.managers.schedule_policy.is_nsa_enable_prefill_cp",
+            "sglang.srt.managers.schedule_policy.is_nsa_prefill_cp_in_seq_split",
             return_value=False,
         )
         self.mock_is_nsa = patcher.start()