From 7e85f2ff9793df75350622ed0c631de01fbfd448 Mon Sep 17 00:00:00 2001
From: pppeng <60355449+ppppeng@users.noreply.github.com>
Date: Sun, 15 Mar 2026 22:19:02 +0800
Subject: [PATCH] [CI] Add test_qwen3_5.py (#7133)

### What this PR does / why we need it?
Add test_qwen3_5.py for base scenarios tp4 on Qwen3.5-27B and
Qwen3.5-35B-A3B.

- vLLM version: main
- vLLM main:
https://github.com/vllm-project/vllm/commit/4034c3d32e30d01639459edd3ab486f56993876d
---------
Signed-off-by: pppeng <zepengliu912@qq.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>
---
 .github/workflows/scripts/config.yaml       |  2 +
 tests/e2e/multicard/4-cards/test_qwen3_5.py | 75 +++++++++++++++++++++
 2 files changed, 77 insertions(+)
 create mode 100644 tests/e2e/multicard/4-cards/test_qwen3_5.py

diff --git a/.github/workflows/scripts/config.yaml b/.github/workflows/scripts/config.yaml
index e3e80e4a..6b7bd2a8 100644
--- a/.github/workflows/scripts/config.yaml
+++ b/.github/workflows/scripts/config.yaml
@@ -143,6 +143,8 @@ e2e-multicard-4-cards:
   # TODO: recover skipped tests
   - name: tests/e2e/multicard/4-cards/test_qwen3_next.py
     estimated_time: 1250
+  - name: tests/e2e/multicard/4-cards/test_qwen3_5.py
+    estimated_time: 60
   - name: tests/e2e/multicard/4-cards/test_data_parallel_tp2.py
     estimated_time: 60
   - name: tests/e2e/multicard/4-cards/test_kimi_k2.py
diff --git a/tests/e2e/multicard/4-cards/test_qwen3_5.py b/tests/e2e/multicard/4-cards/test_qwen3_5.py
new file mode 100644
index 00000000..475086a0
--- /dev/null
+++ b/tests/e2e/multicard/4-cards/test_qwen3_5.py
@@ -0,0 +1,75 @@
+#
+# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
+# Copyright 2023 The vLLM team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# This file is a part of the vllm-ascend project.
+# Adapted from vllm/tests/basic_correctness/test_basic_correctness.py
+#
+from tests.e2e.conftest import VllmRunner
+
+
+def test_qwen3_5_27b_distributed_mp_tp4():
+    example_prompts = [
+        "Hello, my name is",
+    ] * 4
+    max_tokens = 5
+    with VllmRunner("Qwen/Qwen3.5-27B",
+                    tensor_parallel_size=4,
+                    cudagraph_capture_sizes=[1, 2, 4, 8],
+                    max_model_len=4096,
+                    gpu_memory_utilization=0.90,
+                    distributed_executor_backend="mp") as vllm_model:
+        vllm_model.generate_greedy(example_prompts, max_tokens)
+        del vllm_model
+
+
+def test_qwen3_5_35b_distributed_mp_tp4():
+    example_prompts = [
+        "Hello, my name is",
+    ] * 4
+    max_tokens = 5
+    with VllmRunner("Qwen/Qwen3.5-35B-A3B",
+                    tensor_parallel_size=4,
+                    cudagraph_capture_sizes=[1, 2, 4, 8],
+                    max_model_len=4096,
+                    gpu_memory_utilization=0.90,
+                    distributed_executor_backend="mp") as vllm_model:
+        vllm_model.generate_greedy(example_prompts, max_tokens)
+        del vllm_model
+
+
+def test_qwen3_5_35b_distributed_mp_tp4_full_decode_only_mtp3():
+    example_prompts = [
+        "Hello, my name is",
+        "The president of the United States is",
+        "The capital of France is",
+        "The future of AI is",
+    ]
+
+    max_tokens = 20
+    with VllmRunner("Qwen/Qwen3.5-35B-A3B",
+                    tensor_parallel_size=4,
+                    max_model_len=4096,
+                    gpu_memory_utilization=0.90,
+                    distributed_executor_backend="mp",
+                    compilation_config={
+                        "cudagraph_mode": "FULL_DECODE_ONLY",
+                        "cudagraph_capture_sizes": [4, 8, 12, 16],
+                    },
+                    speculative_config={
+                        "method": "qwen3_5_mtp",
+                        "num_speculative_tokens": 3,
+                    }) as vllm_model:
+        vllm_model.generate_greedy(example_prompts, max_tokens)
+        del vllm_model
\ No newline at end of file