[Test] Add e2e test and accuracy test for Qwen3-Next-80B-A3B-Instruct (#3450)

### What this PR does / why we need it?

Add e2e test and accuracy test for Qwen3-Next-80B-A3B-Instruct

### How was this patch tested?
accuracy test:
https://github.com/vllm-project/vllm-ascend/actions/runs/18771221544/job/53556027634?pr=3450
ci test:
https://github.com/vllm-project/vllm-ascend/actions/runs/18771221530/job/53556027614?pr=3450
<img width="1703" height="562" alt="image"
src="https://github.com/user-attachments/assets/973b6cfa-8240-41e3-893a-5024ff8d0693"
/>



- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

Signed-off-by: hfadzxy <[email protected]>

Author: zhangxinyuehfad

.github/workflows/_accuracy_test.yaml

@@ -73,6 +73,16 @@ jobs:
         working-directory: ./vllm-empty
         run: |
           VLLM_TARGET_DEVICE=empty pip install -e .
+        
+      - name: Install Ascend toolkit & triton_ascend (for Qwen3-Next-80B-A3B-Instruct)
+        if: ${{ inputs.model_name == 'Qwen3-Next-80B-A3B-Instruct' }}
+        shell: bash -l {0}
+        run: |
+          wget -q https://vllm-ascend.obs.cn-north-4.myhuaweicloud.com/vllm-ascend/Ascend-BiSheng-toolkit_aarch64.run -O /tmp/Ascend-BiSheng-toolkit_aarch64.run
+          chmod a+x /tmp/Ascend-BiSheng-toolkit_aarch64.run
+          /tmp/Ascend-BiSheng-toolkit_aarch64.run --install
+          . /usr/local/Ascend/8.3.RC1/bisheng_toolkit/set_env.sh
+          python3 -m pip install "https://vllm-ascend.obs.cn-north-4.myhuaweicloud.com/vllm-ascend/triton_ascend-3.2.0.dev20250914-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl"
 
       - name: Resolve vllm-ascend version
         run: |

.github/workflows/accuracy_test.yaml

@@ -57,6 +57,8 @@ jobs:
             model_name: Qwen3-VL-30B-A3B-Instruct
           - runner: a2-2
             model_name: DeepSeek-V2-Lite
+          - runner: a2-4
+            model_name: Qwen3-Next-80B-A3B-Instruct
       fail-fast: false
     # test will be triggered when tag 'accuracy-test' & 'ready-for-test'
     if:  >-

.github/workflows/vllm_ascend_dist.yaml

@@ -19,8 +19,36 @@ name: 'e2e test / a3-test'
 
 on:
   workflow_call:
-
+  push:
+    branches:
+      - 'main'
+      - '*-dev'
+    paths:
+      - '.github/workflows/vllm_ascend_dist.yml'
+      - 'tests/e2e/multicard/**'
+      - 'Dockerfile'
+      - 'vllm_ascend/**'
+      - 'setup.py'
+      - 'pyproject.toml'
+      - 'requirements.txt'
+      - 'cmake/**'
+      - 'CMakeLists.txt'
+      - 'csrc/**'
   pull_request:
+    branches:
+      - 'main'
+      - '*-dev'
+    paths:
+      - '.github/workflows/vllm_ascend_dist.yml'
+      - 'tests/e2e/multicard/**'
+      - 'Dockerfile'
+      - 'vllm_ascend/**'
+      - 'setup.py'
+      - 'pyproject.toml'
+      - 'requirements.txt'
+      - 'cmake/**'
+      - 'CMakeLists.txt'
+      - 'csrc/**'
     types: [ labeled ]
 
 # Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
@@ -39,10 +67,10 @@ concurrency:
 jobs:
   e2e:
     # only trigger e2e test after lint passed and the change is e2e related with pull request.
-    if: ${{ contains(github.event.pull_request.labels.*.name, 'dist-test') && contains(github.event.pull_request.labels.*.name, 'ready-for-test') || github.event_name == 'workflow_dispatch' }}
+    if: ${{ contains(github.event.pull_request.labels.*.name, 'dist-test') && contains(github.event.pull_request.labels.*.name, 'ready-for-test') || github.event_name == 'workflow_dispatch' || github.event_name == 'push'}}
     strategy:
       matrix:
-        os: [linux-aarch64-a3-8]
+        os: [linux-aarch64-a3-4]
         vllm_version: [v0.11.0]
     name: vLLM Ascend test
     runs-on: ${{ matrix.os }}
@@ -61,11 +89,13 @@ jobs:
           sed -i 's|ports.ubuntu.com|mirrors.tuna.tsinghua.edu.cn|g' /etc/apt/sources.list
           pip config set global.index-url https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple
           apt-get update -y
-          apt install git -y
+          apt install git wget curl -y
           git config --global url."https://gh-proxy.test.osinfra.cn/https://github.com/".insteadOf https://github.com/
 
       - name: Checkout vllm-project/vllm-ascend repo
         uses: actions/checkout@v4
+        with:
+          path: ./vllm-ascend
 
       - name: Install system dependencies
         run: |
@@ -85,13 +115,15 @@ jobs:
           VLLM_TARGET_DEVICE=empty pip install -e .
 
       - name: Install vllm-project/vllm-ascend
+        working-directory: ./vllm-ascend
         run: |
           export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi
           export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/x86_64-linux/devlib
           pip install -r requirements-dev.txt
           pip install -v -e .
 
       - name: Run vllm-project/vllm-ascend test for V1 Engine
+        working-directory: ./vllm-ascend
         env:
           VLLM_WORKER_MULTIPROC_METHOD: spawn
           VLLM_USE_MODELSCOPE: True
@@ -102,3 +134,22 @@ jobs:
             tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_DeepSeek_W4A8DYNAMIC \
             tests/e2e/multicard/test_qwen3_moe.py::test_models_distributed_Qwen3_MOE_TP2_WITH_EP \
             tests/e2e/multicard/test_qwen3_moe.py::test_models_distributed_Qwen3_MOE_W8A8_WITH_EP
+
+      - name: Install Ascend toolkit & triton_ascend (for Qwen3-Next-80B-A3B-Instruct)
+        shell: bash -l {0}
+        run: |
+          wget -q https://vllm-ascend.obs.cn-north-4.myhuaweicloud.com/vllm-ascend/Ascend-BiSheng-toolkit_aarch64.run -O /tmp/Ascend-BiSheng-toolkit_aarch64.run
+          chmod a+x /tmp/Ascend-BiSheng-toolkit_aarch64.run
+          /tmp/Ascend-BiSheng-toolkit_aarch64.run --install
+          . /usr/local/Ascend/8.3.RC1/bisheng_toolkit/set_env.sh
+          python3 -m pip install "https://vllm-ascend.obs.cn-north-4.myhuaweicloud.com/vllm-ascend/triton_ascend-3.2.0.dev20250914-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl"
+
+      - name: Run vllm-project/vllm-ascend Qwen3 Next test
+        working-directory: ./vllm-ascend
+        shell: bash -el {0}
+        env:
+          VLLM_WORKER_MULTIPROC_METHOD: spawn
+          VLLM_USE_MODELSCOPE: True
+        run: |
+          . /usr/local/Ascend/8.3.RC1/bisheng_toolkit/set_env.sh
+          pytest -sv tests/e2e/multicard/test_qwen3_next.py

tests/e2e/models/configs/Qwen3-Next-80B-A3B-Instruct.yaml

@@ -0,0 +1,15 @@
+model_name: "Qwen/Qwen3-Next-80B-A3B-Instruct"
+hardware: "Atlas A2 Series"
+model: "vllm"
+tasks:
+- name: "ceval-valid_accountant"
+  metrics:
+  - name: "acc,none"
+    value: 0.98
+max_model_len: 4096
+tensor_parallel_size: 4
+gpu_memory_utilization: 0.7
+enable_expert_parallel: True
+enforce_eager: True
+batch_size: 1
+num_fewshot: 5

tests/e2e/multicard/test_qwen3_next.py

@@ -0,0 +1,38 @@
+#
+# 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
+#
+"""Compare the short outputs of HF and vLLM when using greedy sampling.
+
+Run `pytest tests/e2e/multicard/test_qwen3_next.py`.
+"""
+
+from tests.e2e.conftest import VllmRunner
+
+
+def test_models_distributed_Qwen3_NEXT_TP4():
+    example_prompts = [
+        "Hello, my name is",
+    ]
+    max_tokens = 5
+    with VllmRunner("Qwen/Qwen3-Next-80B-A3B-Instruct",
+                    tensor_parallel_size=4,
+                    max_model_len=4096,
+                    gpu_memory_utilization=0.7,
+                    distributed_executor_backend="mp",
+                    enforce_eager=True) as vllm_model:
+        vllm_model.generate_greedy(example_prompts, max_tokens)