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 16, 07:00 UTC
[CI][CPU] adding numa node number as container name suffix (#10441)
Signed-off-by: Yuan Zhou <yuan.zhou@intel.com>
Yuan committed
Nov 19, 2024 at 21:16 UTC
b4614656b832aa8ac95e5450ca7b861f46049635
1 file changed
+8
-8
.buildkite/run-cpu-test.sh
+8
-8
@@ -13,26 +13,26 @@ numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build -t cpu-test -f Dockerfile.
13
numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --build-arg VLLM_CPU_DISABLE_AVX512="true" -t cpu-test-avx2 -f Dockerfile.cpu .
14
15
# Setup cleanup
16
-remove_docker_container() { docker rm -f cpu-test cpu-test-avx2 || true; }
16
+remove_docker_container() { docker rm -f cpu-test-"$NUMA_NODE" cpu-test-avx2-"$NUMA_NODE" || true; }
17
trap remove_docker_container EXIT
18
remove_docker_container
19
20
# Run the image, setting --shm-size=4g for tensor parallel.
21
docker run -itd --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --cpuset-cpus="$CORE_RANGE" \
22
- --cpuset-mems="$NUMA_NODE" --privileged=true --network host -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test cpu-test
22
+ --cpuset-mems="$NUMA_NODE" --privileged=true --network host -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-"$NUMA_NODE" cpu-test
23
docker run -itd --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --cpuset-cpus="$CORE_RANGE" \
24
- --cpuset-mems="$NUMA_NODE" --privileged=true --network host -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-avx2 cpu-test-avx2
24
+ --cpuset-mems="$NUMA_NODE" --privileged=true --network host -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --shm-size=4g --name cpu-test-avx2-"$NUMA_NODE" cpu-test-avx2
25
26
function cpu_tests() {
27
set -e
28
29
# offline inference
30
- docker exec cpu-test-avx2 bash -c "
30
+ docker exec cpu-test-avx2-"$NUMA_NODE" bash -c "
31
set -e
32
python3 examples/offline_inference.py"
33
34
# Run basic model test
35
- docker exec cpu-test bash -c "
35
+ docker exec cpu-test-"$NUMA_NODE" bash -c "
36
set -e
37
pip install pytest pytest-asyncio \
38
decord einops librosa peft Pillow sentence-transformers soundfile \
@@ -45,20 +45,20 @@ function cpu_tests() {
45
pytest -v -s tests/models/decoder_only/vision_language -m cpu_model"
46
47
# Run compressed-tensor test
48
- docker exec cpu-test bash -c "
48
+ docker exec cpu-test-"$NUMA_NODE" bash -c "
49
set -e
50
pytest -s -v \
51
tests/quantization/test_compressed_tensors.py::test_compressed_tensors_w8a8_static_setup \
52
tests/quantization/test_compressed_tensors.py::test_compressed_tensors_w8a8_dynamic_per_token"
53
54
# Run AWQ test
55
- docker exec cpu-test bash -c "
55
+ docker exec cpu-test-"$NUMA_NODE" bash -c "
56
set -e
57
pytest -s -v \
58
tests/quantization/test_ipex_quant.py"
59
60
# online inference
61
- docker exec cpu-test bash -c "
61
+ docker exec cpu-test-"$NUMA_NODE" bash -c "
62
set -e
63
export VLLM_CPU_KVCACHE_SPACE=10
64
export VLLM_CPU_OMP_THREADS_BIND=$1