louis_lifu
/
AReaL
mirror of https://github.com/inclusionAI/AReaL
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

[Feature] Update Dockerfile to v0.3.0 (#70) 

* update benchmark script

* .

* add benchmark docs

* update dockerfile
This commit is contained in:
Wei Fu 2025-06-03 11:57:56 +08:00 committed by GitHub
parent 2e5728a2c1
commit 326dfe3462
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 66 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

122
Dockerfile
View File

@ -1,76 +1,86 @@
FROM nvcr.io/nvidia/pytorch:24.07-py3 AS v0.1.0 FROM nvcr.io/nvidia/pytorch:25.01-py3
LABEL maintainer="AReaL Team" \
description="AReaL: A Reproducible and Efficient Large Language Model Training Framework" \ WORKDIR /
version="0.1.0"
ENV DEBIAN_FRONTEND=noninteractive ENV DEBIAN_FRONTEND=noninteractive
RUN apt update RUN apt update
RUN apt install -y ca-certificates RUN apt install -y ca-certificates
RUN sed -i "s@http://.*archive.ubuntu.com@https://mirrors.tuna.tsinghua.edu.cn@g" /etc/apt/sources.list RUN sed -i "s@http://.*archive.ubuntu.com@https://mirrors.tuna.tsinghua.edu.cn@g" /etc/apt/sources.list.d/ubuntu.sources
RUN sed -i "s@http://.*security.ubuntu.com@https://mirrors.tuna.tsinghua.edu.cn@g" /etc/apt/sources.list RUN sed -i "s@http://.*security.ubuntu.com@https://mirrors.tuna.tsinghua.edu.cn@g" /etc/apt/sources.list.d/ubuntu.sources
RUN apt update RUN apt update
RUN apt install -y net-tools \ RUN apt install -y net-tools kmod ccache \
libibverbs-dev librdmacm-dev ibverbs-utils \ libibverbs-dev librdmacm-dev ibverbs-utils \
rdmacm-utils python3-pyverbs opensm ibutils perftest rdmacm-utils python3-pyverbs opensm ibutils perftest python3-venv
RUN pip3 install -U pip RUN pip3 install -U pip
RUN pip3 config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple
# set environment variables for building transformer engine ENV NVTE_WITH_USERBUFFERS=1 NVTE_FRAMEWORK=pytorch MPI_HOME=/usr/local/mpi
ENV NVTE_WITH_USERBUFFERS=1 NVTE_FRAMEWORK=pytorch MAX_JOBS=8 MPI_HOME=/usr/local/mpi
ENV PATH="${PATH}:/opt/hpcx/ompi/bin:/opt/hpcx/ucx/bin" ENV PATH="${PATH}:/opt/hpcx/ompi/bin:/opt/hpcx/ucx/bin"
ENV LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/opt/hpcx/ompi/lib:/opt/hpcx/ucx/lib/" ENV LD_LIBRARY_PATH="${LD_LIBRARY_PATH}:/opt/hpcx/ompi/lib:/opt/hpcx/ucx/lib/"
COPY ./requirements.txt /requirements.txt RUN git clone --depth=1 https://github.com/QwenLM/Qwen2.5-Math /qwen2_5-math && mv /qwen2_5-math/evaluation/latex2sympy /latex2sympy && rm -rf /qwen2_5-math \
RUN pip3 install -r /requirements.txt && rm /requirements.txt && python3 -m venv /sympy && \
/sympy/bin/pip config set global.index-url https://pypi.tuna.tsinghua.edu.cn/simple && /sympy/bin/pip config set global.extra-index-url "" \
&& /sympy/bin/pip install /latex2sympy && \
/sympy/bin/pip install regex numpy tqdm datasets python_dateutil sympy==1.12 antlr4-python3-runtime==4.11.1 word2number Pebble timeout-decorator prettytable
# We don't use TransformerEngine's flash-attn integration, so it's okay to disrespect dependencies RUN pip uninstall cugraph-dgl dask-cuda cugraph-service-server raft-dask cugraph cuml \
RUN pip3 install git+https://github.com/NVIDIA/TransformerEngine.git@v1.8 --no-deps --no-build-isolation cugraph-pyg lightning_thunder opt_einsum nvfuser looseversion lightning_utilities -y
RUN pip3 install flash-attn==2.4.2 --no-build-isolation RUN pip3 install -U uv nvidia-ml-py pipdeptree importlib_metadata packaging platformdirs typing_extensions wheel zipp
# Install grouped_gemm for MoE acceleration
RUN pip3 install git+https://github.com/tgale96/grouped_gemm.git@v0.1.4 --no-build-isolation --no-deps
COPY . /AReaL # Things that should be compiled
RUN REAL_CUDA=1 pip3 install -e /AReaL --no-build-isolation ENV MAX_JOBS=64
WORKDIR /AReaL RUN pip install git+https://github.com/fanshiqing/grouped_gemm@v1.1.2
RUN git clone https://github.com/garrett4wade/cugae && pip install -e /cugae --no-build-isolation --verbose
RUN git clone -b v0.11.0 https://github.com/NVIDIA/Megatron-LM.git && \
pip install ./Megatron-LM && rm -rf /Megatron-LM
RUN git clone --depth=1 -b v0.6.3.post1 https://github.com/vllm-project/vllm.git /vllm # flash-attn 2
RUN apt install kmod ccache -y RUN git clone -b v2.6.3 https://github.com/Dao-AILab/flash-attention && \
RUN cd /vllm && \ cd /flash-attention && \
python3 use_existing_torch.py && \ git submodule update --init --recursive && \
pip3 install -r requirements-build.txt && \ pip uninstall -y flash-attn && \
MAX_JOBS=64 pip3 install -e . --no-build-isolation pip install . --no-build-isolation
RUN pip3 install opencv-python-headless==4.5.4.58 # flash-attn 3
RUN mkdir /flash-attn3 && cd /flash-attn3 && git clone -b v2.7.2 https://github.com/Dao-AILab/flash-attention && \
pip install -v /flash-attn3/flash-attention/hopper/ && \
mkdir -p /usr/local/lib/python3.12/dist-packages/flashattn_hopper && \
wget -P /usr/local/lib/python3.12/dist-packages/flashattn_hopper \
https://raw.githubusercontent.com/Dao-AILab/flash-attention/v2.7.2/hopper/flash_attn_interface.py && \
python -c "import flashattn_hopper; import torch; print(torch.__version__)"
RUN apt-get update && apt-get install -y python3.10-venv # sglang depends on flash-infer
ENV TORCH_CUDA_ARCH_LIST="8.0 8.9 9.0 9.0a"
RUN pip install -U setuptools
RUN git clone --recursive -b v0.2.5 https://github.com/flashinfer-ai/flashinfer && \
FLASHINFER_ENABLE_AOT=1 pip install --no-build-isolation --verbose /flashinfer && \
rm -rf /flashinfer
RUN git clone --depth=1 https://github.com/QwenLM/Qwen2.5-Math /qwen2_5-math && mv /qwen2_5-math/evaluation/latex2sympy /latex2sympy # sglang
RUN python3 -m venv /sympy ENV SGL_KERNEL_ENABLE_BF16=1 SGL_KERNEL_ENABLE_FP8=1 SGL_KERNEL_ENABLE_FP4=0
RUN /sympy/bin/pip install /latex2sympy ENV SGL_KERNEL_ENABLE_SM100A=0 SGL_KERNEL_ENABLE_SM90A=1 UV_CONCURRENT_BUILDS=16
RUN /sympy/bin/pip install regex numpy tqdm datasets python_dateutil sympy==1.12 antlr4-python3-runtime==4.11.1 word2number Pebble timeout-decorator prettytable RUN git clone -b v0.4.6.post4 https://github.com/sgl-project/sglang.git && \
FROM v0.1.0 as v0.2.0
LABEL maintainer="AReaL Team" \
description="AReaL: A Reproducible and Efficient Large Language Model Training Framework" \
version="0.2.0"
WORKDIR /
RUN pip uninstall pynvml cugraph-dgl dask-cuda cugraph-service-server raft-dask cugraph cuml cugraph-pyg -y && \
pip install -U six==1.16 transformers==4.48.3 opencv-python-headless==4.7.0.72 \
pipdeptree setuptools importlib_metadata packaging platformdirs \
typing_extensions wheel zipp nvidia-ml-py
ENV TORCH_CUDA_ARCH_LIST="8.0 8.9 9.0 9.0a" FLASHINFER_ENABLE_AOT=1
RUN pip install -v -U git+https://github.com/facebookresearch/xformers.git@v0.0.28.post3#egg=xformers
RUN git clone --recursive -b v0.2.2.post1 https://github.com/flashinfer-ai/flashinfer && \
pip install --no-build-isolation --verbose /flashinfer
RUN git clone -b v0.4.0.post2 https://github.com/sgl-project/sglang.git && \
cd /sglang/sgl-kernel && make build && \ cd /sglang/sgl-kernel && make build && \
pip install /sglang/sgl-kernel/ --force-reinstall --no-build-isolation && \ pip install /sglang/sgl-kernel/ --force-reinstall --no-build-isolation && \
cd /sglang && pip3 install -e "python[all]" cd /sglang && pip3 install -e "python[all]" --no-deps
# sglang dependencies
RUN pip install aiohttp requests tqdm numpy IPython setproctitle \
compressed-tensors datasets decord fastapi hf_transfer huggingface_hub \
interegular "llguidance>=0.7.11,<0.8.0" modelscope ninja orjson packaging \
pillow "prometheus-client>=0.20.0" psutil pydantic pynvml python-multipart \
"pyzmq>=25.1.2" "soundfile==0.13.1" "torchao>=0.7.0" "transformers==4.51.1" \
uvicorn uvloop "xgrammar==0.1.17" cuda-python "outlines>=0.0.44,<=0.1.11" \
partial_json_parser einops jsonlines matplotlib pandas sentence_transformers \
accelerate peft
RUN pip3 install triton==3.1.0 torchao==0.7.0 # vllm, some quantization dependencies required by sglang
RUN git clone -b v0.8.4 --depth=1 https://github.com/vllm-project/vllm.git /vllm && \
git config --global http.version HTTP/1.1 && cd /vllm && \
python3 use_existing_torch.py && \
pip3 install -r requirements/build.txt && \
MAX_JOBS=64 pip install -v . --no-build-isolation && \
rm -rf /vllm
# AReaL and dependencies
RUN git clone https://code.alipay.com/inclusionAI/AReaL && \
pip install -e /AReaL