[CI] reduce npu CI use time and add timeout (#55731)
Signed-off-by: yangzeyu <yangzeyu7@huawei.com>
This commit is contained in:
@@ -6,6 +6,7 @@ steps:
|
|||||||
soft_fail: true
|
soft_fail: true
|
||||||
no_plugin: true
|
no_plugin: true
|
||||||
device: ascend_npu
|
device: ascend_npu
|
||||||
|
timeout_in_minutes: 30
|
||||||
source_file_dependencies:
|
source_file_dependencies:
|
||||||
- vllm/
|
- vllm/
|
||||||
- .buildkite/scripts/hardware_ci/run-npu-test.sh
|
- .buildkite/scripts/hardware_ci/run-npu-test.sh
|
||||||
|
|||||||
@@ -94,18 +94,8 @@ RUN --mount=type=cache,target=/root/.cache/uv \
|
|||||||
|
|
||||||
WORKDIR /workspace/vllm
|
WORKDIR /workspace/vllm
|
||||||
|
|
||||||
# Install vLLM dependencies in advance. Effect: As long as common.txt remains unchanged, the docker cache layer will be valid.
|
|
||||||
COPY requirements/common.txt /workspace/vllm/requirements/common.txt
|
|
||||||
RUN --mount=type=cache,target=/root/.cache/pip \
|
|
||||||
uv pip install -r requirements/common.txt --quiet
|
|
||||||
|
|
||||||
COPY . .
|
COPY . .
|
||||||
|
|
||||||
# Install vLLM
|
|
||||||
RUN --mount=type=cache,target=/root/.cache/uv \
|
|
||||||
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 && \
|
|
||||||
uv pip uninstall -y triton --quiet
|
|
||||||
|
|
||||||
# Install vllm-ascend
|
# Install vllm-ascend
|
||||||
WORKDIR /workspace
|
WORKDIR /workspace
|
||||||
ARG VLLM_ASCEND_REPO=https://github.com/vllm-project/vllm-ascend.git
|
ARG VLLM_ASCEND_REPO=https://github.com/vllm-project/vllm-ascend.git
|
||||||
@@ -113,16 +103,6 @@ ARG VLLM_ASCEND_TAG=main
|
|||||||
RUN git config --global url."https://gh-proxy.test.osinfra.cn/https://github.com/".insteadOf "https://github.com/" && \
|
RUN git config --global url."https://gh-proxy.test.osinfra.cn/https://github.com/".insteadOf "https://github.com/" && \
|
||||||
git clone --depth 1 \$VLLM_ASCEND_REPO --branch \$VLLM_ASCEND_TAG /workspace/vllm-ascend
|
git clone --depth 1 \$VLLM_ASCEND_REPO --branch \$VLLM_ASCEND_TAG /workspace/vllm-ascend
|
||||||
|
|
||||||
# Install vllm dependencies in advance. Effect: As long as common.txt remains unchanged, the docker cache layer will be valid.
|
|
||||||
RUN --mount=type=cache,target=/root/.cache/uv \
|
|
||||||
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
|
|
||||||
|
|
||||||
RUN --mount=type=cache,target=/root/.cache/uv \
|
|
||||||
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
|
|
||||||
source /usr/local/Ascend/nnal/atb/set_env.sh && \
|
|
||||||
export LD_LIBRARY_PATH=\$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/$(uname -i)-linux/devlib && \
|
|
||||||
uv pip install -e /workspace/vllm-ascend/ --extra-index-url https://download.pytorch.org/whl/cpu/ --extra-index-url https://mirrors.huaweicloud.com/ascend/repos/pypi --quiet --index-strategy unsafe-best-match
|
|
||||||
|
|
||||||
ENV VLLM_WORKER_MULTIPROC_METHOD=spawn
|
ENV VLLM_WORKER_MULTIPROC_METHOD=spawn
|
||||||
ENV VLLM_USE_MODELSCOPE=True
|
ENV VLLM_USE_MODELSCOPE=True
|
||||||
|
|
||||||
@@ -187,5 +167,5 @@ docker run \
|
|||||||
"${image_name}" \
|
"${image_name}" \
|
||||||
bash -c '
|
bash -c '
|
||||||
set -e
|
set -e
|
||||||
pytest -v -s tests/e2e/vllm_interface/
|
pytest -v -s tests/e2e/vllm_interface/test_vllm_pr_interface_compatibility.py
|
||||||
'
|
'
|
||||||
|
|||||||
Reference in New Issue
Block a user