Skip to content

Commit 4d1ee72

Browse files
PotabkPotabk
committed
repair e2e test yaml
Signed-off-by: wangli <wangli858794774@gmail.com>
1 parent 7aaa0bf commit 4d1ee72

File tree

1 file changed

+92
-33
lines changed

1 file changed

+92
-33
lines changed

‎.github/workflows/vllm_ascend_test.yaml‎

Lines changed: 92 additions & 33 deletions
Original file line numberDiff line numberDiff line change
@@ -136,9 +136,9 @@ jobs:
136136
strategy:
137137
max-parallel: 2
138138
matrix:
139-
os: [linux-aarch64-a2-1, linux-aarch64-a2-2]
139+
os: [linux-aarch64-a2-1]
140140
vllm_version: [main, v0.10.0]
141-
name: e2e test (${{ matrix.os }} - ${{ matrix.vllm_version }})
141+
name: singlecard e2e test
142142
runs-on: ${{ matrix.os }}
143143
container:
144144
image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
@@ -191,34 +191,93 @@ jobs:
191191
VLLM_WORKER_MULTIPROC_METHOD: spawn
192192
VLLM_USE_MODELSCOPE: True
193193
run: |
194-
if [[ "${{ matrix.os }}" == "linux-aarch64-a2-1" ]]; then
195-
pytest -sv tests/e2e/singlecard/test_offline_inference.py
196-
pytest -sv tests/e2e/singlecard/test_ilama_lora.py
197-
pytest -sv tests/e2e/singlecard/test_guided_decoding.py
198-
pytest -sv tests/e2e/singlecard/test_camem.py
199-
pytest -sv tests/e2e/singlecard/test_embedding.py
200-
pytest -sv tests/e2e/singlecard/ \
201-
--ignore=tests/e2e/singlecard/test_offline_inference.py \
202-
--ignore=tests/e2e/singlecard/test_ilama_lora.py \
203-
--ignore=tests/e2e/singlecard/test_guided_decoding.py \
204-
--ignore=tests/e2e/singlecard/test_camem.py \
205-
--ignore=tests/e2e/singlecard/test_embedding.py \
206-
--ignore=tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py \
207-
--ignore=tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py
208-
# ------------------------------------ v1 spec decode test ------------------------------------ #
209-
VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py
210-
# TODO: revert me when test_v1_spec_decode.py::test_ngram_correctness is fixed
211-
VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py
212-
else
213-
pytest -sv tests/e2e/multicard/test_ilama_lora_tp2.py
214-
# Fixme: run VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py will raise error.
215-
# To avoid oom, we need to run the test in a single process.
216-
pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_DeepSeek_multistream_moe
217-
pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_QwQ
218-
pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_DeepSeek_dbo
219-
pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_DeepSeekV3_dbo
220-
pytest -sv tests/e2e/multicard/test_data_parallel.py
221-
pytest -sv tests/e2e/multicard/ --ignore=tests/e2e/multicard/test_ilama_lora_tp2.py \
222-
--ignore=tests/e2e/multicard/test_offline_inference_distributed.py \
223-
--ignore=tests/e2e/multicard/test_data_parallel.py
224-
fi
194+
pytest -sv tests/e2e/singlecard/test_offline_inference.py
195+
pytest -sv tests/e2e/singlecard/test_ilama_lora.py
196+
pytest -sv tests/e2e/singlecard/test_guided_decoding.py
197+
pytest -sv tests/e2e/singlecard/test_camem.py
198+
pytest -sv tests/e2e/singlecard/test_embedding.py
199+
pytest -sv tests/e2e/singlecard/ \
200+
--ignore=tests/e2e/singlecard/test_offline_inference.py \
201+
--ignore=tests/e2e/singlecard/test_ilama_lora.py \
202+
--ignore=tests/e2e/singlecard/test_guided_decoding.py \
203+
--ignore=tests/e2e/singlecard/test_camem.py \
204+
--ignore=tests/e2e/singlecard/test_embedding.py \
205+
--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
207+
# ------------------------------------ v1 spec decode test ------------------------------------ #
208+
VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/singlecard/spec_decode_v1/test_v1_mtp_correctness.py
209+
# TODO: revert me when test_v1_spec_decode.py::test_ngram_correctness is fixed
210+
VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/singlecard/spec_decode_v1/test_v1_spec_decode.py
211+
212+
e2e-4-cards:
213+
needs: [e2e]
214+
if: ${{ needs.e2e.result == 'success' }}
215+
strategy:
216+
max-parallel: 2
217+
matrix:
218+
os: [linux-aarch64-a2-2]
219+
vllm_version: [main, v0.10.0]
220+
name: multicard e2e test
221+
runs-on: ${{ matrix.os }}
222+
container:
223+
image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
224+
env:
225+
VLLM_LOGGING_LEVEL: ERROR
226+
VLLM_USE_MODELSCOPE: True
227+
steps:
228+
- name: Check npu and CANN info
229+
run: |
230+
npu-smi info
231+
cat /usr/local/Ascend/ascend-toolkit/latest/"$(uname -i)"-linux/ascend_toolkit_install.info
232+
233+
- name: Config mirrors
234+
run: |
235+
sed -Ei 's@(ports|archive).ubuntu.com@cache-service.nginx-pypi-cache.svc.cluster.local:8081@g' /etc/apt/sources.list
236+
pip config set global.index-url http://cache-service.nginx-pypi-cache.svc.cluster.local/pypi/simple
237+
pip config set global.trusted-host cache-service.nginx-pypi-cache.svc.cluster.local
238+
apt-get update -y
239+
apt install git -y
240+
241+
- name: Checkout vllm-project/vllm-ascend repo
242+
uses: actions/checkout@v4
243+
244+
- name: Install system dependencies
245+
run: |
246+
apt-get -y install `cat packages.txt`
247+
apt-get -y install gcc g++ cmake libnuma-dev
248+
249+
- name: Checkout vllm-project/vllm repo
250+
uses: actions/checkout@v4
251+
with:
252+
repository: vllm-project/vllm
253+
ref: ${{ matrix.vllm_version }}
254+
path: ./vllm-empty
255+
256+
- name: Install vllm-project/vllm from source
257+
working-directory: ./vllm-empty
258+
run: |
259+
VLLM_TARGET_DEVICE=empty pip install -e .
260+
261+
- name: Install vllm-project/vllm-ascend
262+
env:
263+
PIP_EXTRA_INDEX_URL: https://mirrors.huaweicloud.com/ascend/repos/pypi
264+
run: |
265+
pip install -r requirements-dev.txt
266+
pip install -v -e .
267+
268+
- name: Run vllm-project/vllm-ascend test
269+
env:
270+
VLLM_WORKER_MULTIPROC_METHOD: spawn
271+
VLLM_USE_MODELSCOPE: True
272+
run: |
273+
pytest -sv tests/e2e/multicard/test_ilama_lora_tp2.py
274+
# Fixme: run VLLM_USE_MODELSCOPE=True pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py will raise error.
275+
# To avoid oom, we need to run the test in a single process.
276+
pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_DeepSeek_multistream_moe
277+
pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_QwQ
278+
pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_DeepSeek_dbo
279+
pytest -sv tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_DeepSeekV3_dbo
280+
pytest -sv tests/e2e/multicard/test_data_parallel.py
281+
pytest -sv tests/e2e/multicard/ --ignore=tests/e2e/multicard/test_ilama_lora_tp2.py \
282+
--ignore=tests/e2e/multicard/test_offline_inference_distributed.py \
283+
--ignore=tests/e2e/multicard/test_data_parallel.py

0 commit comments

Comments
 (0)