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, 20:45 UTC
[CI/Build] Cross python wheel (#6394)
Robert Shaw committed
Jul 14, 2024 at 21:54 UTC
a754dc2cb964c78fd9da514aa0b87a0241e97d1f
2 files changed
+4
-38
.buildkite/release-pipeline.yaml
+2
-36
@@ -1,51 +1,17 @@
1
steps:
2
- - label: "Build wheel default - Python {{matrix.python_version}}, CUDA {{matrix.cuda_version}}"
2
+ - label: "Build wheel - CUDA {{matrix.cuda_version}}"
3
agents:
4
queue: cpu_queue
5
commands:
6
- - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg CUDA_VERSION={{matrix.cuda_version}} --build-arg PYTHON_VERSION={{matrix.python_version}} --tag vllm-ci:build-image --target build --progress plain ."
6
+ - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg CUDA_VERSION={{matrix.cuda_version}} --tag vllm-ci:build-image --target build --progress plain ."
7
- "mkdir artifacts"
8
- "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
9
# rename the files to change linux -> manylinux1
10
- "for f in artifacts/dist/*.whl; do mv -- \"$$f\" \"$${f/linux/manylinux1}\"; done"
11
- "aws s3 cp --recursive artifacts/dist s3://vllm-wheels/$BUILDKITE_COMMIT/"
12
- "aws s3 cp --recursive artifacts/dist s3://vllm-wheels/nightly/"
13
- matrix:
14
- setup:
15
- cuda_version:
16
- - "12.1.0"
17
- python_version:
18
- - "3.10"
19
- - "3.11"
20
-
21
- - block: "Build wheels full"
22
-
23
- - label: "Build wheel - Python {{matrix.python_version}}, CUDA {{matrix.cuda_version}}"
24
- agents:
25
- queue: cpu_queue
26
- commands:
27
- - "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg CUDA_VERSION={{matrix.cuda_version}} --build-arg PYTHON_VERSION={{matrix.python_version}} --tag vllm-ci:build-image --target build --progress plain ."
28
- - "mkdir artifacts"
29
- - "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
30
- - "for f in artifacts/dist/*.whl; do mv -- \"$$f\" \"$${f/linux/manylinux1}\"; done"
31
- - "aws s3 cp --recursive artifacts/dist s3://vllm-wheels/$BUILDKITE_COMMIT/"
32
- - "aws s3 cp --recursive artifacts/dist s3://vllm-wheels/nightly/"
13
matrix:
14
setup:
15
cuda_version:
16
- "11.8.0"
17
- "12.1.0"
38
- python_version:
39
- - "3.8"
40
- - "3.9"
41
- - "3.10"
42
- - "3.11"
43
- adjustments:
44
- - with:
45
- cuda_version: "12.1.0"
46
- python_version: "3.10"
47
- skip: true
48
- - with:
49
- cuda_version: "12.1.0"
50
- python_version: "3.11"
51
- skip: true
Dockerfile
+2
-2
@@ -104,7 +104,7 @@ RUN --mount=type=cache,target=/root/.cache/pip \
104
&& export SCCACHE_REGION=us-west-2 \
105
&& export CMAKE_BUILD_TYPE=Release \
106
&& sccache --show-stats \
107
- && python3 setup.py bdist_wheel --dist-dir=dist \
107
+ && python3 setup.py bdist_wheel --dist-dir=dist --py-limited-api=cp38 \
108
&& sccache --show-stats; \
109
fi
110
@@ -112,7 +112,7 @@ ENV CCACHE_DIR=/root/.cache/ccache
112
RUN --mount=type=cache,target=/root/.cache/ccache \
113
--mount=type=cache,target=/root/.cache/pip \
114
if [ "$USE_SCCACHE" != "1" ]; then \
115
- python3 setup.py bdist_wheel --dist-dir=dist; \
115
+ python3 setup.py bdist_wheel --dist-dir=dist --py-limited-api=cp38; \
116
fi
117
118
# check the size of the wheel, we cannot upload wheels larger than 100MB