Skip to content

Commit 6874d66

Browse files
zhangxinyuehfadzhangxinyuehfad
authored
[CI]Add e2e test for 310p (vllm-project#1879)
### What this PR does / why we need it? Add e2e test for 310p: trigger conditions：tag, labels(ready-for-test, e2e-310p-test), schedule image: m.daocloud.io/quay.io/ascend/cann:8.1.rc1-310p-ubuntu22.04-py3.10 runner: linux-aarch64-310p-1, linux-aarch64-310p-4 model: IntervitensInc/pangu-pro-moe-model, Qwen/Qwen3-0.6B-Base, Qwen/Qwen2.5-7B-Instruct - vLLM version: v0.10.0 - vLLM main: vllm-project/vllm@b917da4 Signed-off-by: hfadzxy <[email protected]>
1 parent 34dd24a commit 6874d66

File tree

5 files changed

+230
-2
lines changed

5 files changed

+230
-2
lines changed

‎.github/actionlint.yaml‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -7,4 +7,7 @@ self-hosted-runner:
77
- linux-aarch64-a2-4
88
- linux-aarch64-a2-8
99
- linux-arm64-npu-static-8
10+
- linux-aarch64-310p-1
11+
- linux-aarch64-310p-2
12+
- linux-aarch64-310p-4
1013
- ubuntu-24.04-arm

‎.github/workflows/vllm_ascend_test.yaml‎

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -203,7 +203,8 @@ jobs:
203203
--ignore=tests/e2e/singlecard/test_camem.py \
204204
--ignore=tests/e2e/singlecard/test_embedding.py \
205205
--ignore=tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py \
206-
--ignore=tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py
206+
--ignore=tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py \
207+
--ignore=tests/e2e/singlecard/test_offline_inference_310p.py
207208
# ------------------------------------ v1 spec decode test ------------------------------------ #
208209
VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py
209210
# TODO: revert me when test_v1_spec_decode.py::test_ngram_correctness is fixed
@@ -280,4 +281,5 @@ jobs:
280281
pytest -sv tests/e2e/multicard/test_data_parallel.py
281282
pytest -sv tests/e2e/multicard/ --ignore=tests/e2e/multicard/test_ilama_lora_tp2.py \
282283
--ignore=tests/e2e/multicard/test_offline_inference_distributed.py \
283-
--ignore=tests/e2e/multicard/test_data_parallel.py
284+
--ignore=tests/e2e/multicard/test_data_parallel.py \
285+
--ignore=tests/e2e/multicard/test_offline_inference_310p.py

‎.github/workflows/vllm_ascend_test_310p.yaml‎

Lines changed: 117 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,117 @@
1+
#
2+
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
3+
#
4+
# Licensed under the Apache License, Version 2.0 (the "License");
5+
# you may not use this file except in compliance with the License.
6+
# You may obtain a copy of the License at
7+
#
8+
# http://www.apache.org/licenses/LICENSE-2.0
9+
#
10+
# Unless required by applicable law or agreed to in writing, software
11+
# distributed under the License is distributed on an "AS IS" BASIS,
12+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13+
# See the License for the specific language governing permissions and
14+
# limitations under the License.
15+
# This file is a part of the vllm-ascend project.
16+
#
17+
18+
name: 'e2e test / 310p-test'
19+
20+
on:
21+
push:
22+
tags:
23+
- 'v*'
24+
schedule:
25+
# Runs every 6 hours
26+
- cron: '0 */6 * * *'
27+
pull_request:
28+
types: [ labeled ]
29+
30+
# Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
31+
# declared as "shell: bash -el {0}" on steps that need to be properly activated.
32+
# It's used to activate ascend-toolkit environment variables.
33+
defaults:
34+
run:
35+
shell: bash -el {0}
36+
37+
# only cancel in-progress runs of the same workflow
38+
# and ignore the lint / 1 card / 4 cards test type
39+
concurrency:
40+
group: ${{ github.workflow }}-${{ github.ref }}
41+
cancel-in-progress: true
42+
43+
jobs:
44+
e2e:
45+
# e2e-310p-test will be triggered when tag 'e2e-310p-test' & 'ready-for-test' or schedule job
46+
if: >-
47+
${{
48+
(contains(github.event.pull_request.labels.*.name, 'e2e-310p-test')) &&
49+
contains(github.event.pull_request.labels.*.name, 'ready-for-test') ||
50+
github.event_name == 'schedule' || github.event_name == 'push'
51+
}}
52+
strategy:
53+
max-parallel: 2
54+
matrix:
55+
os: [linux-aarch64-310p-1, linux-aarch64-310p-4]
56+
vllm_version: [main, v0.10.0]
57+
name: 310p e2e test
58+
runs-on: ${{ matrix.os }}
59+
container:
60+
# TODO(yikun): Remove m.daocloud.io prefix when infra proxy ready
61+
image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.1.rc1-310p-ubuntu22.04-py3.10
62+
env:
63+
VLLM_LOGGING_LEVEL: ERROR
64+
VLLM_USE_MODELSCOPE: True
65+
steps:
66+
- name: Check npu and CANN info
67+
run: |
68+
npu-smi info
69+
cat /usr/local/Ascend/ascend-toolkit/latest/"$(uname -i)"-linux/ascend_toolkit_install.info
70+
71+
- name: Config mirrors
72+
run: |
73+
sed -Ei 's@(ports|archive)[email protected]:8081@g' /etc/apt/sources.list
74+
pip config set global.index-url http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple
75+
pip config set global.trusted-host cache-service.nginx-pypi-cache.svc.cluster.local
76+
apt-get update -y
77+
apt install git -y
78+
79+
- name: Checkout vllm-project/vllm-ascend repo
80+
uses: actions/checkout@v4
81+
82+
- name: Install system dependencies
83+
run: |
84+
apt-get -y install `cat packages.txt`
85+
apt-get -y install git vim wget net-tools gcc g++ cmake libnuma-dev curl gnupg2
86+
87+
- name: Checkout vllm-project/vllm repo
88+
uses: actions/checkout@v4
89+
with:
90+
repository: vllm-project/vllm
91+
ref: ${{ matrix.vllm_version }}
92+
path: ./vllm-empty
93+
94+
- name: Install vllm-project/vllm from source
95+
working-directory: ./vllm-empty
96+
run: |
97+
VLLM_TARGET_DEVICE=empty pip install -e .
98+
99+
- name: Install vllm-project/vllm-ascend
100+
run: |
101+
export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi
102+
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/x86_64-linux/devlib
103+
export SOC_VERSION=ASCEND310P3
104+
pip install -r requirements-dev.txt
105+
pip install -v -e .
106+
107+
- name: Run e2e test
108+
env:
109+
VLLM_WORKER_MULTIPROC_METHOD: spawn
110+
VLLM_USE_MODELSCOPE: True
111+
PYTORCH_NPU_ALLOC_CONF: max_split_size_mb:256
112+
run: |
113+
if [[ "${{ matrix.os }}" == "linux-aarch64-310p-1" ]]; then
114+
pytest -sv tests/e2e/singlecard/test_offline_inference_310p.py
115+
else
116+
pytest -sv tests/e2e/multicard/test_offline_inference_310p.py
117+
fi

‎tests/e2e/multicard/test_offline_inference_310p.py‎

Lines changed: 62 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,62 @@
1+
#
2+
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
3+
# Copyright 2023 The vLLM team.
4+
#
5+
# Licensed under the Apache License, Version 2.0 (the "License");
6+
# you may not use this file except in compliance with the License.
7+
# You may obtain a copy of the License at
8+
#
9+
# http://www.apache.org/licenses/LICENSE-2.0
10+
#
11+
# Unless required by applicable law or agreed to in writing, software
12+
# distributed under the License is distributed on an "AS IS" BASIS,
13+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14+
# See the License for the specific language governing permissions and
15+
# limitations under the License.
16+
# This file is a part of the vllm-ascend project.
17+
import pytest
18+
import vllm # noqa: F401
19+
20+
import vllm_ascend # noqa: F401
21+
from tests.e2e.conftest import VllmRunner
22+
23+
# Pangu local model path
24+
MODELS = [
25+
"IntervitensInc/pangu-pro-moe-model",
26+
]
27+
# set additional config for ascend scheduler and torchair graph
28+
ADDITIONAL_CONFIG = [{
29+
"additional_config": {
30+
"torchair_graph_config": {
31+
"enabled": True
32+
},
33+
"ascend_scheduler_config": {
34+
"enabled": True,
35+
}
36+
}
37+
}]
38+
39+
40+
@pytest.mark.parametrize("model", MODELS)
41+
@pytest.mark.parametrize("dtype", ["float16"])
42+
@pytest.mark.parametrize("max_tokens", [5])
43+
@pytest.mark.parametrize("enfore_eager", [True, False])
44+
@pytest.mark.parametrize("additional_config", ADDITIONAL_CONFIG)
45+
def test_pangu_model(model: str, dtype: str, max_tokens: int,
46+
enfore_eager: bool, additional_config: dict) -> None:
47+
if enfore_eager:
48+
additional_config = {}
49+
example_prompts = [
50+
"Hello, my name is",
51+
"The future of AI is",
52+
]
53+
54+
with VllmRunner(model,
55+
tensor_parallel_size=4,
56+
dtype=dtype,
57+
max_model_len=1024,
58+
enforce_eager=True,
59+
enable_expert_parallel=True,
60+
additional_config=additional_config,
61+
distributed_executor_backend="mp") as vllm_model:
62+
vllm_model.generate_greedy(example_prompts, max_tokens)

‎tests/e2e/singlecard/test_offline_inference_310p.py‎

Lines changed: 44 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,44 @@
1+
#
2+
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
3+
# Copyright 2023 The vLLM team.
4+
#
5+
# Licensed under the Apache License, Version 2.0 (the "License");
6+
# you may not use this file except in compliance with the License.
7+
# You may obtain a copy of the License at
8+
#
9+
# http://www.apache.org/licenses/LICENSE-2.0
10+
#
11+
# Unless required by applicable law or agreed to in writing, software
12+
# distributed under the License is distributed on an "AS IS" BASIS,
13+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14+
# See the License for the specific language governing permissions and
15+
# limitations under the License.
16+
# This file is a part of the vllm-ascend project.
17+
import pytest
18+
import vllm # noqa: F401
19+
20+
import vllm_ascend # noqa: F401
21+
from tests.e2e.conftest import VllmRunner
22+
23+
MODELS = ["Qwen/Qwen3-0.6B-Base", "Qwen/Qwen2.5-7B-Instruct"]
24+
25+
26+
@pytest.mark.parametrize("model", MODELS)
27+
@pytest.mark.parametrize("dtype", ["float16"])
28+
@pytest.mark.parametrize("max_tokens", [5])
29+
def test_models(model: str, dtype: str, max_tokens: int) -> None:
30+
example_prompts = [
31+
"Hello, my name is",
32+
"The future of AI is",
33+
]
34+
35+
with VllmRunner(model,
36+
tensor_parallel_size=1,
37+
dtype=dtype,
38+
max_model_len=2048,
39+
enforce_eager=True,
40+
compilation_config={
41+
"custom_ops":
42+
["none", "+rms_norm", "+rotary_embedding"]
43+
}) as vllm_model:
44+
vllm_model.generate_greedy(example_prompts, max_tokens)

0 commit comments

Comments
 (0)