[CI] reduce npu CI use time and add timeout (#55731)
Signed-off-by: yangzeyu <yangzeyu7@huawei.com>
Zeyu Yang committed
Sep 7, 2026 at 22:51 UTC
ad2f18e8af18f43af932e08a3dfb7c8c068aebc2
2 files changed
+2
-21
.buildkite/hardware_tests/ascend_npu.yaml
+1
@@ -6,6 +6,7 @@ steps:
6
soft_fail: true
7
no_plugin: true
8
device: ascend_npu
9
+ timeout_in_minutes: 30
10
source_file_dependencies:
11
- vllm/
12
- .buildkite/scripts/hardware_ci/run-npu-test.sh
.buildkite/scripts/hardware_ci/run-npu-test.sh
+1
-21
@@ -94,18 +94,8 @@ RUN --mount=type=cache,target=/root/.cache/uv \
94
95
WORKDIR /workspace/vllm
96
97
-# Install vLLM dependencies in advance. Effect: As long as common.txt remains unchanged, the docker cache layer will be valid.
98
-COPY requirements/common.txt /workspace/vllm/requirements/common.txt
99
-RUN --mount=type=cache,target=/root/.cache/pip \
100
- uv pip install -r requirements/common.txt --quiet
101
-
97
COPY . .
98
104
-# Install vLLM
105
-RUN --mount=type=cache,target=/root/.cache/uv \
106
- VLLM_TARGET_DEVICE="empty" uv pip install -e /workspace/vllm/ --extra-index-url https://download.pytorch.org/whl/cpu/ --quiet --index-strategy unsafe-best-match && \
107
- uv pip uninstall -y triton --quiet
108
-
99
# Install vllm-ascend
100
WORKDIR /workspace
101
ARG VLLM_ASCEND_REPO=https://github.com/vllm-project/vllm-ascend.git
@@ -113,16 +103,6 @@ ARG VLLM_ASCEND_TAG=main
103
RUN git config --global url."https://gh-proxy.test.osinfra.cn/https://github.com/".insteadOf "https://github.com/" && \
104
git clone --depth 1 \$VLLM_ASCEND_REPO --branch \$VLLM_ASCEND_TAG /workspace/vllm-ascend
105
116
-# Install vllm dependencies in advance. Effect: As long as common.txt remains unchanged, the docker cache layer will be valid.
117
-RUN --mount=type=cache,target=/root/.cache/uv \
118
- uv pip install -r /workspace/vllm-ascend/requirements.txt --extra-index-url https://mirrors.huaweicloud.com/ascend/repos/pypi --quiet --index-strategy unsafe-best-match
119
-
120
-RUN --mount=type=cache,target=/root/.cache/uv \
121
- source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
122
- source /usr/local/Ascend/nnal/atb/set_env.sh && \
123
- export LD_LIBRARY_PATH=\$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/$(uname -i)-linux/devlib && \
124
- uv pip install -e /workspace/vllm-ascend/ --extra-index-url https://download.pytorch.org/whl/cpu/ --extra-index-url https://mirrors.huaweicloud.com/ascend/repos/pypi --quiet --index-strategy unsafe-best-match
125
-
106
ENV VLLM_WORKER_MULTIPROC_METHOD=spawn
107
ENV VLLM_USE_MODELSCOPE=True
108
@@ -187,5 +167,5 @@ docker run \
167
"${image_name}" \
168
bash -c '
169
set -e
190
- pytest -v -s tests/e2e/vllm_interface/
170
+ pytest -v -s tests/e2e/vllm_interface/test_vllm_pr_interface_compatibility.py
171
'