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 15, 03:30 UTC
[CI][Intel Gaudi][vllm-Plugin]Add CI for hpu-plugin-v1-test (#20196)
Signed-off-by: Chendi Xue <chendi.xue@intel.com>
Chendi.Xue committed
Jun 30, 2025 at 23:17 UTC
a2f14dc8f9bb04bd782d1aa4d2e6364841d63d6c
1 file changed
+40
-8
.buildkite/scripts/hardware_ci/run-hpu-test.sh
+40
-8
@@ -2,10 +2,34 @@
2
3
# This script build the CPU docker image and run the offline inference inside the container.
4
# It serves a sanity check for compilation and basic model usage.
5
-set -ex
5
+set -exuo pipefail
6
7
# Try building the docker image
8
-docker build -t hpu-test-env -f docker/Dockerfile.hpu .
8
+cat <<EOF | docker build -t hpu-plugin-v1-test-env -f - .
9
+FROM 1.22-413-pt2.7.1:latest
10
+
11
+COPY ./ /workspace/vllm
12
+
13
+WORKDIR /workspace/vllm
14
+
15
+RUN pip install -v -r requirements/hpu.txt
16
+RUN pip install git+https://github.com/vllm-project/vllm-gaudi.git
17
+
18
+ENV no_proxy=localhost,127.0.0.1
19
+ENV PT_HPU_ENABLE_LAZY_COLLECTIVES=true
20
+
21
+RUN VLLM_TARGET_DEVICE=hpu python3 setup.py install
22
+
23
+# install development dependencies (for testing)
24
+RUN python3 -m pip install -e tests/vllm_test_utils
25
+
26
+WORKDIR /workspace/
27
+
28
+RUN git clone https://github.com/vllm-project/vllm-gaudi.git
29
+
30
+RUN ln -s /workspace/vllm/tests && ln -s /workspace/vllm/examples && ln -s /workspace/vllm/benchmarks
31
+
32
+EOF
33
34
# Setup cleanup
35
# certain versions of HPU software stack have a bug that can
@@ -14,13 +38,21 @@ docker build -t hpu-test-env -f docker/Dockerfile.hpu .
38
# functions, while other platforms only need one remove_docker_container
39
# function.
40
EXITCODE=1
17
-remove_docker_containers() { docker rm -f hpu-test || true; docker rm -f hpu-test-tp2 || true; }
18
-remove_docker_containers_and_exit() { remove_docker_containers; exit $EXITCODE; }
19
-trap remove_docker_containers_and_exit EXIT
41
+remove_docker_containers() { docker rm -f hpu-plugin-v1-test || true; }
42
+trap 'remove_docker_containers; exit $EXITCODE;' EXIT
43
remove_docker_containers
44
22
-# Run the image and launch offline inference
23
-docker run --runtime=habana --name=hpu-test --network=host -e HABANA_VISIBLE_DEVICES=all -e VLLM_SKIP_WARMUP=true --entrypoint="" hpu-test-env python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m
24
-docker run --runtime=habana --name=hpu-test-tp2 --network=host -e HABANA_VISIBLE_DEVICES=all -e VLLM_SKIP_WARMUP=true --entrypoint="" hpu-test-env python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --tensor-parallel-size 2
45
+echo "Running HPU plugin v1 test"
46
+docker run --rm --runtime=habana --name=hpu-plugin-v1-test --network=host \
47
+ -e HABANA_VISIBLE_DEVICES=all \
48
+ hpu-plugin-v1-test-env \
49
+ /bin/bash "/workspace/vllm-gaudi/tests/upstream_tests/ci_tests.sh"
50
51
EXITCODE=$?
52
+if [ $EXITCODE -eq 0 ]; then
53
+ echo "Test with basic model passed"
54
+else
55
+ echo "Test with basic model FAILED with exit code: $EXITCODE" >&2
56
+fi
57
+
58
+# The trap will handle the container removal and final exit.
\ No newline at end of file