Dockerfile 3.11 KB
Newer Older
Simon Mo's avatar
Simon Mo committed
1
2
3
4
# The vLLM Dockerfile is used to construct vLLM image that can be directly used
# to run the OpenAI compatible server.

#################### BASE BUILD IMAGE ####################
5
FROM nvidia/cuda:12.1.0-devel-ubuntu22.04 AS dev
Stephen Krider's avatar
Stephen Krider committed
6
7

RUN apt-get update -y \
Simon Mo's avatar
Simon Mo committed
8
    && apt-get install -y python3-pip git
Stephen Krider's avatar
Stephen Krider committed
9
10
11
12
13
14
15

WORKDIR /workspace

# install build and runtime dependencies
COPY requirements.txt requirements.txt
RUN --mount=type=cache,target=/root/.cache/pip \
    pip install -r requirements.txt
16

Stephen Krider's avatar
Stephen Krider committed
17
18
19
20
# install development dependencies
COPY requirements-dev.txt requirements-dev.txt
RUN --mount=type=cache,target=/root/.cache/pip \
    pip install -r requirements-dev.txt
Simon Mo's avatar
Simon Mo committed
21
22
#################### BASE BUILD IMAGE ####################

Stephen Krider's avatar
Stephen Krider committed
23

Simon Mo's avatar
Simon Mo committed
24
#################### EXTENSION BUILD IMAGE ####################
Stephen Krider's avatar
Stephen Krider committed
25
26
FROM dev AS build

27
28
29
30
31
# install build dependencies
COPY requirements-build.txt requirements-build.txt
RUN --mount=type=cache,target=/root/.cache/pip \
    pip install -r requirements-build.txt

Stephen Krider's avatar
Stephen Krider committed
32
33
34
35
36
37
38
# copy input files
COPY csrc csrc
COPY setup.py setup.py
COPY requirements.txt requirements.txt
COPY pyproject.toml pyproject.toml
COPY vllm/__init__.py vllm/__init__.py

Simon Mo's avatar
Simon Mo committed
39
# cuda arch list used by torch
40
41
ARG torch_cuda_arch_list='7.0 7.5 8.0 8.6 8.9 9.0+PTX'
ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
Stephen Krider's avatar
Stephen Krider committed
42
# max jobs used by Ninja to build extensions
43
44
ARG max_jobs=2
ENV MAX_JOBS=${max_jobs}
45
46
47
# number of threads used by nvcc
ARG nvcc_threads=8
ENV NVCC_THREADS=$nvcc_threads
48
49
# make sure punica kernels are built (for LoRA)
ENV VLLM_INSTALL_PUNICA_KERNELS=1
50

Stephen Krider's avatar
Stephen Krider committed
51
RUN python3 setup.py build_ext --inplace
Simon Mo's avatar
Simon Mo committed
52
#################### EXTENSION Build IMAGE ####################
Stephen Krider's avatar
Stephen Krider committed
53

Simon Mo's avatar
Simon Mo committed
54
55

#################### TEST IMAGE ####################
Stephen Krider's avatar
Stephen Krider committed
56
57
58
59
60
# image to run unit testing suite
FROM dev AS test

# copy pytorch extensions separately to avoid having to rebuild
# when python code changes
Simon Mo's avatar
Simon Mo committed
61
62
63
64
65
66
67
68
WORKDIR /vllm-workspace
# ADD is used to preserve directory structure
ADD . /vllm-workspace/
COPY --from=build /workspace/vllm/*.so /vllm-workspace/vllm/
# ignore build dependencies installation because we are using pre-complied extensions
RUN rm pyproject.toml
RUN --mount=type=cache,target=/root/.cache/pip VLLM_USE_PRECOMPILED=1 pip install . --verbose
#################### TEST IMAGE ####################
Stephen Krider's avatar
Stephen Krider committed
69
70


Simon Mo's avatar
Simon Mo committed
71
#################### RUNTIME BASE IMAGE ####################
Stephen Krider's avatar
Stephen Krider committed
72
# use CUDA base as CUDA runtime dependencies are already installed via pip
73
FROM nvidia/cuda:12.1.0-base-ubuntu22.04 AS vllm-base
Stephen Krider's avatar
Stephen Krider committed
74
75
76
77
78
79
80
81
82

# libnccl required for ray
RUN apt-get update -y \
    && apt-get install -y python3-pip

WORKDIR /workspace
COPY requirements.txt requirements.txt
RUN --mount=type=cache,target=/root/.cache/pip \
    pip install -r requirements.txt
Simon Mo's avatar
Simon Mo committed
83
#################### RUNTIME BASE IMAGE ####################
Stephen Krider's avatar
Stephen Krider committed
84
85


Simon Mo's avatar
Simon Mo committed
86
#################### OPENAI API SERVER ####################
Stephen Krider's avatar
Stephen Krider committed
87
88
# openai api server alternative
FROM vllm-base AS vllm-openai
89
# install additional dependencies for openai api server
Stephen Krider's avatar
Stephen Krider committed
90
RUN --mount=type=cache,target=/root/.cache/pip \
91
    pip install accelerate
Stephen Krider's avatar
Stephen Krider committed
92

93
COPY --from=build /workspace/vllm/*.so /workspace/vllm/
94
COPY vllm vllm
Stephen Krider's avatar
Stephen Krider committed
95
96

ENTRYPOINT ["python3", "-m", "vllm.entrypoints.openai.api_server"]
Simon Mo's avatar
Simon Mo committed
97
#################### OPENAI API SERVER ####################