A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 9, 07:30 UTC
[CI][HPU]accelerate hpu test by skip python re-install and clean container name (#33286)
Signed-off-by: Chendi Xue <chendi.xue@intel.com>
Chendi.Xue committed
Jan 30, 2026 at 15:36 UTC
2b465570e6dd327e8422ef9c87e9b2b1454ceaed
1 file changed
+14
-6
.buildkite/scripts/hardware_ci/run-hpu-test.sh
+14
-6
@@ -5,7 +5,9 @@
5
set -exuo pipefail
6
7
# Try building the docker image
8
-cat <<EOF | docker build -t hpu-plugin-v1-test-env -f - .
8
+image_name="hpu/upstream-vllm-ci:${BUILDKITE_COMMIT}"
9
+container_name="hpu-upstream-vllm-ci-${BUILDKITE_COMMIT}-container"
10
+cat <<EOF | docker build -t ${image_name} -f - .
11
FROM gaudi-base-image:latest
12
13
COPY ./ /workspace/vllm
@@ -15,7 +17,8 @@ WORKDIR /workspace/vllm
17
ENV no_proxy=localhost,127.0.0.1
18
ENV PT_HPU_ENABLE_LAZY_COLLECTIVES=true
19
18
-RUN VLLM_TARGET_DEVICE=empty pip install .
20
+RUN bash -c 'pip install -r <(sed "/^torch/d" requirements/build.txt)'
21
+RUN VLLM_TARGET_DEVICE=empty pip install --no-build-isolation -e .
22
RUN pip install git+https://github.com/vllm-project/vllm-gaudi.git
23
24
# install development dependencies (for testing)
@@ -36,15 +39,20 @@ EOF
39
# functions, while other platforms only need one remove_docker_container
40
# function.
41
EXITCODE=1
39
-remove_docker_containers() { docker rm -f hpu-plugin-v1-test || true; }
42
+remove_docker_containers() { docker rm -f ${container_name} || true; }
43
trap 'remove_docker_containers; exit $EXITCODE;' EXIT
44
remove_docker_containers
45
46
echo "Running HPU plugin v1 test"
44
-docker run --rm --runtime=habana --name=hpu-plugin-v1-test --network=host \
47
+docker run --rm --runtime=habana --name=${container_name} --network=host \
48
-e HABANA_VISIBLE_DEVICES=all \
46
- hpu-plugin-v1-test-env \
47
- /bin/bash "/workspace/vllm-gaudi/tests/upstream_tests/ci_tests.sh"
49
+ -e VLLM_SKIP_WARMUP=true \
50
+ -e PT_HPU_ENABLE_LAZY_COLLECTIVES=true \
51
+ -e PT_HPU_LAZY_MODE=1 \
52
+ "${image_name}" \
53
+ /bin/bash -c '
54
+ cd vllm; timeout 120s python -u examples/offline_inference/basic/generate.py --model facebook/opt-125m
55
+'
56
57
EXITCODE=$?
58
if [ $EXITCODE -eq 0 ]; then