update dockerfile
This commit is contained in:
Your Name
@@ -166,300 +166,301 @@ cd ${STAGE_DIR}
|
|||||||
rm -rf ${STAGE_DIR}/MLNX_OFED_LINUX-${MLNX_OFED_VERSION}-ubuntu22.04-x86_64*
|
rm -rf ${STAGE_DIR}/MLNX_OFED_LINUX-${MLNX_OFED_VERSION}-ubuntu22.04-x86_64*
|
||||||
EOT
|
EOT
|
||||||
|
|
||||||
# ENV NV_PEER_MEM_VERSION=1.2
|
ENV NV_PEER_MEM_VERSION=1.2
|
||||||
# ENV NV_PEER_MEM_TAG=${NV_PEER_MEM_VERSION}-0
|
ENV NV_PEER_MEM_TAG=${NV_PEER_MEM_VERSION}-0
|
||||||
|
RUN <<EOT
|
||||||
|
#!/bin/bash
|
||||||
|
source /opt/conda/etc/profile.d/conda.sh
|
||||||
|
conda activate ${CONDA_ENV_NAME}
|
||||||
|
# install nv_peer_mem
|
||||||
|
rm -rf ${STAGE_DIR}
|
||||||
|
mkdir -p ${STAGE_DIR}
|
||||||
|
git clone https://github.com/Mellanox/nv_peer_memory.git --branch ${NV_PEER_MEM_TAG} ${STAGE_DIR}/nv_peer_memory
|
||||||
|
cd ${STAGE_DIR}/nv_peer_memory
|
||||||
|
./build_module.sh
|
||||||
|
cd ${STAGE_DIR}
|
||||||
|
tar xzf ${STAGE_DIR}/nvidia-peer-memory_${NV_PEER_MEM_VERSION}.orig.tar.gz
|
||||||
|
cd ${STAGE_DIR}/nvidia-peer-memory-${NV_PEER_MEM_VERSION}
|
||||||
|
apt-get update
|
||||||
|
apt --fix-broken install -y
|
||||||
|
apt-get install -y dkms
|
||||||
|
dpkg-buildpackage -us -uc
|
||||||
|
dpkg -i ${STAGE_DIR}/nvidia-peer-memory_${NV_PEER_MEM_TAG}_all.deb
|
||||||
|
EOT
|
||||||
|
|
||||||
|
# install mpi
|
||||||
|
ENV PATH=/usr/local/mpi/bin:${PATH}
|
||||||
|
ENV LD_LIBRARY_PATH=/usr/local/lib:/usr/local/mpi/lib:/usr/local/mpi/lib64:${LD_LIBRARY_PATH}
|
||||||
|
RUN <<EOT
|
||||||
|
#!/bin/bash
|
||||||
|
source /opt/conda/etc/profile.d/conda.sh
|
||||||
|
conda activate ${CONDA_ENV_NAME}
|
||||||
|
# OPENMPI
|
||||||
|
rm -rf ${STAGE_DIR}
|
||||||
|
mkdir -p ${STAGE_DIR}
|
||||||
|
cd ${STAGE_DIR}
|
||||||
|
wget -q -O - https://download.open-mpi.org/release/open-mpi/v${OPENMPI_BASEVERSION}/openmpi-${OPENMPI_VERSION}.tar.gz | tar xzf -
|
||||||
|
cd openmpi-${OPENMPI_VERSION}
|
||||||
|
./configure --prefix=/usr/local/openmpi-${OPENMPI_VERSION}
|
||||||
|
make -j"$(nproc)" install
|
||||||
|
ln -s /usr/local/openmpi-${OPENMPI_VERSION} /usr/local/mpi
|
||||||
|
# Sanity check:
|
||||||
|
test -f /usr/local/mpi/bin/mpic++
|
||||||
|
cd ${STAGE_DIR}
|
||||||
|
rm -r ${STAGE_DIR}/openmpi-${OPENMPI_VERSION}
|
||||||
|
# Create a wrapper for OpenMPI to allow running as root by default
|
||||||
|
mv /usr/local/mpi/bin/mpirun /usr/local/mpi/bin/mpirun.real
|
||||||
|
echo '#!/bin/bash' > /usr/local/mpi/bin/mpirun
|
||||||
|
echo 'mpirun.real --allow-run-as-root --prefix /usr/local/mpi "$@"' >> /usr/local/mpi/bin/mpirun
|
||||||
|
chmod a+x /usr/local/mpi/bin/mpirun
|
||||||
|
EOT
|
||||||
|
|
||||||
|
# SSH daemon port inside container cannot conflict with host OS port
|
||||||
|
# ENV SSH_PORT=2222
|
||||||
# RUN <<EOT
|
# RUN <<EOT
|
||||||
# #!/bin/bash
|
# #!/bin/bash
|
||||||
# source /opt/conda/etc/profile.d/conda.sh
|
# source /opt/conda/etc/profile.d/conda.sh
|
||||||
# conda activate ${CONDA_ENV_NAME}
|
# conda activate ${CONDA_ENV_NAME}
|
||||||
# # install nv_peer_mem
|
# cat /etc/ssh/sshd_config > ${STAGE_DIR}/sshd_config && \
|
||||||
# rm -rf ${STAGE_DIR}
|
# sed "0,/^Port 22/s//Port ${SSH_PORT}/" ${STAGE_DIR}/sshd_config > /etc/ssh/sshd_config
|
||||||
# mkdir -p ${STAGE_DIR}
|
# EOT
|
||||||
# git clone https://github.com/Mellanox/nv_peer_memory.git --branch ${NV_PEER_MEM_TAG} ${STAGE_DIR}/nv_peer_memory
|
|
||||||
# cd ${STAGE_DIR}/nv_peer_memory
|
# 29.78 Usage: install.sh [options...]
|
||||||
# ./build_module.sh
|
# 29.78
|
||||||
# cd ${STAGE_DIR}
|
# 29.78 By default will install deepspeed and all third party dependencies across all machines listed in
|
||||||
# tar xzf ${STAGE_DIR}/nvidia-peer-memory_${NV_PEER_MEM_VERSION}.orig.tar.gz
|
# 29.78 hostfile (hostfile: /job/hostfile). If no hostfile exists, will only install locally
|
||||||
# cd ${STAGE_DIR}/nvidia-peer-memory-${NV_PEER_MEM_VERSION}
|
# 29.78
|
||||||
|
# 29.78 [optional]
|
||||||
|
# 29.78 -l, --local_only Install only on local machine
|
||||||
|
# 29.78 -s, --pip_sudo Run pip install with sudo (default: no sudo)
|
||||||
|
# 29.78 -r, --allow_sudo Allow script to be run by root (probably don't want this, instead use --pip_sudo)
|
||||||
|
# 29.78 -n, --no_clean Do not clean prior build state, by default prior build files are removed before building wheels
|
||||||
|
# 29.78 -m, --pip_mirror Use the specified pip mirror (default: the default pip mirror)
|
||||||
|
# 29.78 -H, --hostfile Path to MPI-style hostfile (default: /job/hostfile)
|
||||||
|
# 29.78 -e, --examples Checkout deepspeed example submodule (no install)
|
||||||
|
# 29.78 -v, --verbose Verbose logging
|
||||||
|
# 29.78 -h, --help This help text
|
||||||
|
|
||||||
|
RUN <<EOT
|
||||||
|
#!/bin/bash
|
||||||
|
source /opt/conda/etc/profile.d/conda.sh
|
||||||
|
conda activate ${CONDA_ENV_NAME}
|
||||||
|
useradd --create-home --uid 1000 --shell /bin/bash deepspeed
|
||||||
|
usermod -aG sudo deepspeed
|
||||||
|
echo "deepspeed ALL=(ALL) NOPASSWD: ALL" >> /etc/sudoers
|
||||||
|
EOT
|
||||||
|
|
||||||
|
# install cutlass https://github.com/NVIDIA/cutlass
|
||||||
|
# H100: architecture is Hopper (cutlass need add : cmake .. -DCUTLASS_NVCC_ARCHS="90a" )
|
||||||
|
# A100: architecture is Ampere
|
||||||
|
# V100: architecture is Volta
|
||||||
|
# T4: architecture is Turing
|
||||||
|
# ENV CUDACXX=${CUDA_INSTALL_PATH}/bin/nvcc
|
||||||
|
# 70：适用于 NVIDIA Volta 架构（如 Tesla V100）。
|
||||||
|
# 75：适用于 NVIDIA Turing 架构（如 Tesla T4）。
|
||||||
|
# 80：适用于 NVIDIA Ampere 架构（如 A100）。
|
||||||
|
# 90a：适用于 NVIDIA Hopper 架构（如 H100）。
|
||||||
|
# 89:GeForce RTX 4090
|
||||||
|
ARG DCUTLASS_NVCC_ARCHS="80;89;90a"
|
||||||
|
ENV DCUTLASS_NVCC_ARCHS=${DCUTLASS_NVCC_ARCHS}
|
||||||
|
RUN <<EOT
|
||||||
|
#!/bin/bash
|
||||||
|
source /opt/conda/etc/profile.d/conda.sh
|
||||||
|
conda activate ${CONDA_ENV_NAME}
|
||||||
|
git clone https://github.com/NVIDIA/cutlass /opt/cutlass
|
||||||
|
cd /opt/cutlass
|
||||||
|
git checkout .
|
||||||
|
git checkout master
|
||||||
|
mkdir build
|
||||||
|
cd build
|
||||||
|
cmake .. -DCUTLASS_NVCC_ARCHS=${DCUTLASS_NVCC_ARCHS} -DCUTLASS_ENABLE_TESTS=OFF -DCUTLASS_UNITY_BUILD_ENABLED=ON # compiles for NVIDIA Hopper GPU architecture, like H100
|
||||||
|
make -j"$(nproc)" install
|
||||||
|
cd ..
|
||||||
|
# make test_unit -j"$(nproc)"
|
||||||
|
# make test_unit_gemm_warp -j"$(nproc)"
|
||||||
|
EOT
|
||||||
|
|
||||||
|
# Some Packages from https://github.com/microsoft/DeepSpeed/blob/master/docker/Dockerfile
|
||||||
|
# RUN <<EOT
|
||||||
|
# source /opt/conda/etc/profile.d/conda.sh
|
||||||
|
# conda activate ${CONDA_ENV_NAME}
|
||||||
# apt-get update
|
# apt-get update
|
||||||
# apt-get install -y dkms
|
# apt-get install -y --no-install-recommends libsndfile-dev libcupti-dev libjpeg-dev libpng-dev screen libaio-dev
|
||||||
# dpkg-buildpackage -us -uc
|
# python -m pip install pipdeptree \
|
||||||
# dpkg -i ${STAGE_DIR}/nvidia-peer-memory_${NV_PEER_MEM_TAG}_all.deb
|
# psutil \
|
||||||
|
# yappi \
|
||||||
|
# cffi \
|
||||||
|
# ipdb \
|
||||||
|
# pandas \
|
||||||
|
# matplotlib \
|
||||||
|
# py3nvml \
|
||||||
|
# pyarrow \
|
||||||
|
# graphviz \
|
||||||
|
# astor \
|
||||||
|
# boto3 \
|
||||||
|
# tqdm \
|
||||||
|
# sentencepiece \
|
||||||
|
# msgpack \
|
||||||
|
# requests \
|
||||||
|
# pandas \
|
||||||
|
# sphinx \
|
||||||
|
# sphinx_rtd_theme \
|
||||||
|
# scipy \
|
||||||
|
# numpy \
|
||||||
|
# scikit-learn \
|
||||||
|
# nvidia-ml-py3 \
|
||||||
|
# mpi4py
|
||||||
# EOT
|
# EOT
|
||||||
|
|
||||||
# # install mpi
|
# install deepspeed step 1
|
||||||
# ENV PATH=/usr/local/mpi/bin:${PATH}
|
RUN <<EOT
|
||||||
# ENV LD_LIBRARY_PATH=/usr/local/lib:/usr/local/mpi/lib:/usr/local/mpi/lib64:${LD_LIBRARY_PATH}
|
#!/bin/bash
|
||||||
# RUN <<EOT
|
source /opt/conda/etc/profile.d/conda.sh
|
||||||
# #!/bin/bash
|
conda activate ${CONDA_ENV_NAME}
|
||||||
# source /opt/conda/etc/profile.d/conda.sh
|
/opt/conda/envs/${CONDA_ENV_NAME}/bin/python -m pip install setuptools==${SETUPTOOLS_VERSION}
|
||||||
# conda activate ${CONDA_ENV_NAME}
|
# install oneapi for deepspeed
|
||||||
# # OPENMPI
|
git clone https://github.com/oneapi-src/oneCCL.git ${STAGE_DIR}/oneCCL
|
||||||
# rm -rf ${STAGE_DIR}
|
cd ${STAGE_DIR}/oneCCL
|
||||||
# mkdir -p ${STAGE_DIR}
|
git checkout .
|
||||||
# cd ${STAGE_DIR}
|
git checkout master
|
||||||
# wget -q -O - https://download.open-mpi.org/release/open-mpi/v${OPENMPI_BASEVERSION}/openmpi-${OPENMPI_VERSION}.tar.gz | tar xzf -
|
mkdir build
|
||||||
# cd openmpi-${OPENMPI_VERSION}
|
cd build
|
||||||
# ./configure --prefix=/usr/local/openmpi-${OPENMPI_VERSION}
|
cmake .. -DCMAKE_INSTALL_PREFIX=/usr/local
|
||||||
# make -j"$(nproc)" install
|
make -j"$(nproc)" install
|
||||||
# ln -s /usr/local/openmpi-${OPENMPI_VERSION} /usr/local/mpi
|
EOT
|
||||||
# # Sanity check:
|
|
||||||
# test -f /usr/local/mpi/bin/mpic++
|
|
||||||
# cd ${STAGE_DIR}
|
|
||||||
# rm -r ${STAGE_DIR}/openmpi-${OPENMPI_VERSION}
|
|
||||||
# # Create a wrapper for OpenMPI to allow running as root by default
|
|
||||||
# mv /usr/local/mpi/bin/mpirun /usr/local/mpi/bin/mpirun.real
|
|
||||||
# echo '#!/bin/bash' > /usr/local/mpi/bin/mpirun
|
|
||||||
# echo 'mpirun.real --allow-run-as-root --prefix /usr/local/mpi "$@"' >> /usr/local/mpi/bin/mpirun
|
|
||||||
# chmod a+x /usr/local/mpi/bin/mpirun
|
|
||||||
# EOT
|
|
||||||
|
|
||||||
# # SSH daemon port inside container cannot conflict with host OS port
|
# install deepspeed step 2
|
||||||
# # ENV SSH_PORT=2222
|
ARG CUDA_ARCH_LIST="80;86;89;90"
|
||||||
# # RUN <<EOT
|
ENV CUDA_ARCH_LIST=${CUDA_ARCH_LIST}
|
||||||
# # #!/bin/bash
|
RUN <<EOT
|
||||||
# # source /opt/conda/etc/profile.d/conda.sh
|
#!/bin/bash
|
||||||
# # conda activate ${CONDA_ENV_NAME}
|
source /opt/conda/etc/profile.d/conda.sh
|
||||||
# # cat /etc/ssh/sshd_config > ${STAGE_DIR}/sshd_config && \
|
conda activate ${CONDA_ENV_NAME}
|
||||||
# # sed "0,/^Port 22/s//Port ${SSH_PORT}/" ${STAGE_DIR}/sshd_config > /etc/ssh/sshd_config
|
git clone https://github.com/microsoft/DeepSpeed-Kernels.git ${STAGE_DIR}/DeepSpeed-Kernels
|
||||||
# # EOT
|
cd ${STAGE_DIR}/DeepSpeed-Kernels
|
||||||
|
# CUDA_ARCH_LIST=${CUDA_ARCH_LIST} python setup.py bdist_wheel
|
||||||
|
# pip install dist/deepspeed_kernels-*.whl
|
||||||
|
CUDA_ARCH_LIST=${CUDA_ARCH_LIST} python -m pip install -v .
|
||||||
|
EOT
|
||||||
|
|
||||||
# # 29.78 Usage: install.sh [options...]
|
ARG DEEPSPEED_VERSION="v0.14.3"
|
||||||
# # 29.78
|
ENV DEEPSPEED_VERSION=${DEEPSPEED_VERSION}
|
||||||
# # 29.78 By default will install deepspeed and all third party dependencies across all machines listed in
|
ARG DEEPSPEED_INSTALL_FLAGS="--allow_sudo --pip_sudo --verbose"
|
||||||
# # 29.78 hostfile (hostfile: /job/hostfile). If no hostfile exists, will only install locally
|
ENV DEEPSPEED_INSTALL_FLAGS=${DEEPSPEED_INSTALL_FLAGS}
|
||||||
# # 29.78
|
ARG DS_BUILD_SPARSE_ATTN=0
|
||||||
# # 29.78 [optional]
|
ENV DS_BUILD_SPARSE_ATTN=${DS_BUILD_SPARSE_ATTN}
|
||||||
# # 29.78 -l, --local_only Install only on local machine
|
ARG DS_BUILD_FUSED_ADAM=1
|
||||||
# # 29.78 -s, --pip_sudo Run pip install with sudo (default: no sudo)
|
ENV DS_BUILD_FUSED_ADAM=${DS_BUILD_FUSED_ADAM}
|
||||||
# # 29.78 -r, --allow_sudo Allow script to be run by root (probably don't want this, instead use --pip_sudo)
|
ARG DS_BUILD_CPU_ADAM=1
|
||||||
# # 29.78 -n, --no_clean Do not clean prior build state, by default prior build files are removed before building wheels
|
ENV DS_BUILD_CPU_ADAM=${DS_BUILD_CPU_ADAM}
|
||||||
# # 29.78 -m, --pip_mirror Use the specified pip mirror (default: the default pip mirror)
|
ARG DS_BUILD_OPS=1
|
||||||
# # 29.78 -H, --hostfile Path to MPI-style hostfile (default: /job/hostfile)
|
ENV DS_BUILD_OPS=${DS_BUILD_OPS}
|
||||||
# # 29.78 -e, --examples Checkout deepspeed example submodule (no install)
|
ARG HOSTFILE_CONTENT=""
|
||||||
# # 29.78 -v, --verbose Verbose logging
|
ENV HOSTFILE_CONTENT=${HOSTFILE_CONTENT}
|
||||||
# # 29.78 -h, --help This help text
|
ENV CUTLASS_PATH='/opt/cutlass'
|
||||||
|
ENV CUDA_HOME='/usr/local/cuda'
|
||||||
|
ENV LD_LIBRARY_PATH=${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}
|
||||||
|
ENV PATH=${CUDA_HOME}/bin:${PATH}
|
||||||
|
# install deepspeed step 3
|
||||||
|
RUN <<EOT
|
||||||
|
#!/bin/bash
|
||||||
|
source /opt/conda/etc/profile.d/conda.sh
|
||||||
|
conda activate ${CONDA_ENV_NAME}
|
||||||
|
git clone https://github.com/microsoft/DeepSpeed.git ${STAGE_DIR}/DeepSpeed
|
||||||
|
cd ${STAGE_DIR}/DeepSpeed
|
||||||
|
git checkout ${DEEPSPEED_VERSION}
|
||||||
|
sed 's/pip install/python -m pip install/' install.sh > install_modified.sh
|
||||||
|
chmod +x ./install_modified.sh
|
||||||
|
# 检查 HOSTFILE_CONTENT 并写入文件
|
||||||
|
if [ -n "${HOSTFILE_CONTENT}" ]; then
|
||||||
|
echo "${HOSTFILE_CONTENT}" > /tmp/hostfile
|
||||||
|
INSTALL_CMD="./install_modified.sh ${DEEPSPEED_INSTALL_FLAGS} --hostfile /tmp/hostfile"
|
||||||
|
else
|
||||||
|
INSTALL_CMD="./install_modified.sh ${DEEPSPEED_INSTALL_FLAGS}"
|
||||||
|
fi
|
||||||
|
eval $INSTALL_CMD
|
||||||
|
# compile deepspeed ops
|
||||||
|
cat <<EOF > ~/compile_deepspeed_ops.py
|
||||||
|
import deepspeed
|
||||||
|
|
||||||
# RUN <<EOT
|
def compile_ops():
|
||||||
# #!/bin/bash
|
builders = [
|
||||||
# source /opt/conda/etc/profile.d/conda.sh
|
deepspeed.ops.op_builder.AsyncIOBuilder,
|
||||||
# conda activate ${CONDA_ENV_NAME}
|
deepspeed.ops.op_builder.FusedAdamBuilder,
|
||||||
# useradd --create-home --uid 1000 --shell /bin/bash deepspeed
|
deepspeed.ops.op_builder.CPUAdamBuilder,
|
||||||
# usermod -aG sudo deepspeed
|
deepspeed.ops.op_builder.CPUAdagradBuilder,
|
||||||
# echo "deepspeed ALL=(ALL) NOPASSWD: ALL" >> /etc/sudoers
|
deepspeed.ops.op_builder.CPULionBuilder,
|
||||||
# EOT
|
deepspeed.ops.op_builder.EvoformerAttnBuilder,
|
||||||
|
deepspeed.ops.op_builder.FPQuantizerBuilder,
|
||||||
|
deepspeed.ops.op_builder.FusedLambBuilder,
|
||||||
|
deepspeed.ops.op_builder.FusedLionBuilder,
|
||||||
|
deepspeed.ops.op_builder.QuantizerBuilder,
|
||||||
|
deepspeed.ops.op_builder.RaggedOpsBuilder,
|
||||||
|
deepspeed.ops.op_builder.RandomLTDBuilder,
|
||||||
|
deepspeed.ops.op_builder.SparseAttnBuilder,
|
||||||
|
deepspeed.ops.op_builder.SpatialInferenceBuilder,
|
||||||
|
deepspeed.ops.op_builder.TransformerBuilder,
|
||||||
|
deepspeed.ops.op_builder.StochasticTransformerBuilder,
|
||||||
|
]
|
||||||
|
|
||||||
# # install cutlass https://github.com/NVIDIA/cutlass
|
for builder in builders:
|
||||||
# # H100: architecture is Hopper (cutlass need add : cmake .. -DCUTLASS_NVCC_ARCHS="90a" )
|
print(f"Compiling {builder.__name__}")
|
||||||
# # A100: architecture is Ampere
|
builder().load()
|
||||||
# # V100: architecture is Volta
|
|
||||||
# # T4: architecture is Turing
|
|
||||||
# # ENV CUDACXX=${CUDA_INSTALL_PATH}/bin/nvcc
|
|
||||||
# # 70：适用于 NVIDIA Volta 架构（如 Tesla V100）。
|
|
||||||
# # 75：适用于 NVIDIA Turing 架构（如 Tesla T4）。
|
|
||||||
# # 80：适用于 NVIDIA Ampere 架构（如 A100）。
|
|
||||||
# # 90a：适用于 NVIDIA Hopper 架构（如 H100）。
|
|
||||||
# # 89:GeForce RTX 4090
|
|
||||||
# ARG DCUTLASS_NVCC_ARCHS="80;89;90a"
|
|
||||||
# ENV DCUTLASS_NVCC_ARCHS=${DCUTLASS_NVCC_ARCHS}
|
|
||||||
# RUN <<EOT
|
|
||||||
# #!/bin/bash
|
|
||||||
# source /opt/conda/etc/profile.d/conda.sh
|
|
||||||
# conda activate ${CONDA_ENV_NAME}
|
|
||||||
# git clone https://github.com/NVIDIA/cutlass /opt/cutlass
|
|
||||||
# cd /opt/cutlass
|
|
||||||
# git checkout .
|
|
||||||
# git checkout master
|
|
||||||
# mkdir build
|
|
||||||
# cd build
|
|
||||||
# cmake .. -DCUTLASS_NVCC_ARCHS=${DCUTLASS_NVCC_ARCHS} -DCUTLASS_ENABLE_TESTS=OFF -DCUTLASS_UNITY_BUILD_ENABLED=ON # compiles for NVIDIA Hopper GPU architecture, like H100
|
|
||||||
# make -j"$(nproc)" install
|
|
||||||
# cd ..
|
|
||||||
# # make test_unit -j"$(nproc)"
|
|
||||||
# # make test_unit_gemm_warp -j"$(nproc)"
|
|
||||||
# EOT
|
|
||||||
|
|
||||||
# # Some Packages from https://github.com/microsoft/DeepSpeed/blob/master/docker/Dockerfile
|
if __name__ == "__main__":
|
||||||
# # RUN <<EOT
|
compile_ops()
|
||||||
# # source /opt/conda/etc/profile.d/conda.sh
|
EOF
|
||||||
# # conda activate ${CONDA_ENV_NAME}
|
python compile_deepspeed_ops.py
|
||||||
# # apt-get update
|
ds_report
|
||||||
# # apt-get install -y --no-install-recommends libsndfile-dev libcupti-dev libjpeg-dev libpng-dev screen libaio-dev
|
# clean up
|
||||||
# # python -m pip install pipdeptree \
|
# rm -f deepspeed/git_version_info_installed.py
|
||||||
# # psutil \
|
# rm -rf dist build deepspeed.egg-info
|
||||||
# # yappi \
|
# python setup.py bdist_wheel
|
||||||
# # cffi \
|
# DS_BUILD_OPS=${DS_BUILD_OPS} pip install -v dist/deepspeed*.whl
|
||||||
# # ipdb \
|
# DS_BUILD_OPS=${DS_BUILD_OPS} pip install -v -r requirements/requirements.txt
|
||||||
# # pandas \
|
# pip install numpy==1.22.4 # ImportError: cannot import name 'BUFSIZE' from 'numpy' (/opt/conda/envs/deepspeed/lib/python3.10/site-packages/numpy/__init__.py) wait for fix in numpy=2.0.0
|
||||||
# # matplotlib \
|
EOT
|
||||||
# # py3nvml \
|
|
||||||
# # pyarrow \
|
|
||||||
# # graphviz \
|
|
||||||
# # astor \
|
|
||||||
# # boto3 \
|
|
||||||
# # tqdm \
|
|
||||||
# # sentencepiece \
|
|
||||||
# # msgpack \
|
|
||||||
# # requests \
|
|
||||||
# # pandas \
|
|
||||||
# # sphinx \
|
|
||||||
# # sphinx_rtd_theme \
|
|
||||||
# # scipy \
|
|
||||||
# # numpy \
|
|
||||||
# # scikit-learn \
|
|
||||||
# # nvidia-ml-py3 \
|
|
||||||
# # mpi4py
|
|
||||||
# # EOT
|
|
||||||
|
|
||||||
# # install deepspeed step 1
|
# install transformers and flash-attn
|
||||||
# RUN <<EOT
|
RUN <<EOT
|
||||||
# #!/bin/bash
|
#!/bin/bash
|
||||||
# source /opt/conda/etc/profile.d/conda.sh
|
source /opt/conda/etc/profile.d/conda.sh
|
||||||
# conda activate ${CONDA_ENV_NAME}
|
conda activate ${CONDA_ENV_NAME}
|
||||||
# /opt/conda/envs/${CONDA_ENV_NAME}/bin/python -m pip install setuptools==${SETUPTOOLS_VERSION}
|
# install transformers
|
||||||
# # install oneapi for deepspeed
|
git clone https://github.com/huggingface/transformers ${STAGE_DIR}/transformers
|
||||||
# git clone https://github.com/oneapi-src/oneCCL.git ${STAGE_DIR}/oneCCL
|
cd ${STAGE_DIR}/transformers
|
||||||
# cd ${STAGE_DIR}/oneCCL
|
python3 ./setup.py develop
|
||||||
# git checkout .
|
python3 -m pip install -U --no-cache-dir "pydantic<2"
|
||||||
# git checkout master
|
# install flash-attn
|
||||||
# mkdir build
|
# pip install packaging -i https://pypi.org/simple/ --trusted-host pypi.org
|
||||||
# cd build
|
pip install flash-attn --no-build-isolation -i https://pypi.org/simple/ --trusted-host pypi.org
|
||||||
# cmake .. -DCMAKE_INSTALL_PREFIX=/usr/local
|
EOT
|
||||||
# make -j"$(nproc)" install
|
|
||||||
# EOT
|
|
||||||
|
|
||||||
# # install deepspeed step 2
|
# other packages
|
||||||
# ARG CUDA_ARCH_LIST="80;86;89;90"
|
ENV TORCH_CUDA_ARCH_LIST="80;86;89;90"
|
||||||
# ENV CUDA_ARCH_LIST=${CUDA_ARCH_LIST}
|
RUN <<EOT
|
||||||
# RUN <<EOT
|
#!/bin/bash
|
||||||
# #!/bin/bash
|
source /opt/conda/etc/profile.d/conda.sh
|
||||||
# source /opt/conda/etc/profile.d/conda.sh
|
conda activate ${CONDA_ENV_NAME}
|
||||||
# conda activate ${CONDA_ENV_NAME}
|
pip3 install optimum
|
||||||
# git clone https://github.com/microsoft/DeepSpeed-Kernels.git ${STAGE_DIR}/DeepSpeed-Kernels
|
pip3 install peft tiktoken \
|
||||||
# cd ${STAGE_DIR}/DeepSpeed-Kernels
|
tqdm matplotlib seaborn numpy pandas scikit-learn diffusers \
|
||||||
# # CUDA_ARCH_LIST=${CUDA_ARCH_LIST} python setup.py bdist_wheel
|
huggingface_hub spacy blobfile pycocotools \
|
||||||
# # pip install dist/deepspeed_kernels-*.whl
|
open_clip_torch \
|
||||||
# CUDA_ARCH_LIST=${CUDA_ARCH_LIST} python -m pip install -v .
|
zstandard -i https://pypi.org/simple/ --trusted-host pypi.org
|
||||||
# EOT
|
EOT
|
||||||
|
|
||||||
# ARG DEEPSPEED_VERSION="v0.14.3"
|
ARG DEEPSPEED_TRAIN='/data/train_data'
|
||||||
# ENV DEEPSPEED_VERSION=${DEEPSPEED_VERSION}
|
ENV DEEPSPEED_TRAIN=DEEPSPEED_TRAIN
|
||||||
# ARG DEEPSPEED_INSTALL_FLAGS="--allow_sudo --pip_sudo --verbose"
|
ARG DEEPSPEED_VALIDATION='/data/validation_data'
|
||||||
# ENV DEEPSPEED_INSTALL_FLAGS=${DEEPSPEED_INSTALL_FLAGS}
|
ENV DEEPSPEED_VALIDATION=DEEPSPEED_VALIDATION
|
||||||
# ARG DS_BUILD_SPARSE_ATTN=0
|
ARG NCCL_SOCKET_IFNAME='eth0'
|
||||||
# ENV DS_BUILD_SPARSE_ATTN=${DS_BUILD_SPARSE_ATTN}
|
|
||||||
# ARG DS_BUILD_FUSED_ADAM=1
|
|
||||||
# ENV DS_BUILD_FUSED_ADAM=${DS_BUILD_FUSED_ADAM}
|
|
||||||
# ARG DS_BUILD_CPU_ADAM=1
|
|
||||||
# ENV DS_BUILD_CPU_ADAM=${DS_BUILD_CPU_ADAM}
|
|
||||||
# ARG DS_BUILD_OPS=1
|
|
||||||
# ENV DS_BUILD_OPS=${DS_BUILD_OPS}
|
|
||||||
# ARG HOSTFILE_CONTENT=""
|
|
||||||
# ENV HOSTFILE_CONTENT=${HOSTFILE_CONTENT}
|
|
||||||
# ENV CUTLASS_PATH='/opt/cutlass'
|
|
||||||
# ENV CUDA_HOME='/usr/local/cuda'
|
|
||||||
# ENV LD_LIBRARY_PATH=${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}
|
|
||||||
# ENV PATH=${CUDA_HOME}/bin:${PATH}
|
|
||||||
# # install deepspeed step 3
|
|
||||||
# RUN <<EOT
|
|
||||||
# #!/bin/bash
|
|
||||||
# source /opt/conda/etc/profile.d/conda.sh
|
|
||||||
# conda activate ${CONDA_ENV_NAME}
|
|
||||||
# git clone https://github.com/microsoft/DeepSpeed.git ${STAGE_DIR}/DeepSpeed
|
|
||||||
# cd ${STAGE_DIR}/DeepSpeed
|
|
||||||
# git checkout ${DEEPSPEED_VERSION}
|
|
||||||
# sed 's/pip install/python -m pip install/' install.sh > install_modified.sh
|
|
||||||
# chmod +x ./install_modified.sh
|
|
||||||
# # 检查 HOSTFILE_CONTENT 并写入文件
|
|
||||||
# if [ -n "${HOSTFILE_CONTENT}" ]; then
|
|
||||||
# echo "${HOSTFILE_CONTENT}" > /tmp/hostfile
|
|
||||||
# INSTALL_CMD="./install_modified.sh ${DEEPSPEED_INSTALL_FLAGS} --hostfile /tmp/hostfile"
|
|
||||||
# else
|
|
||||||
# INSTALL_CMD="./install_modified.sh ${DEEPSPEED_INSTALL_FLAGS}"
|
|
||||||
# fi
|
|
||||||
# eval $INSTALL_CMD
|
|
||||||
# # compile deepspeed ops
|
|
||||||
# cat <<EOF > ~/compile_deepspeed_ops.py
|
|
||||||
# import deepspeed
|
|
||||||
|
|
||||||
# def compile_ops():
|
# RUN echo 'export CUDA_HOME=/usr/local/cuda' >> ~/.bashrc && \
|
||||||
# builders = [
|
# echo 'export PATH=${CUDA_HOME}/bin:${PATH}' >> ~/.bashrc && \
|
||||||
# deepspeed.ops.op_builder.AsyncIOBuilder,
|
# echo 'export CUTLASS_PATH=/opt/cutlass' >> ~/.bashrc && \
|
||||||
# deepspeed.ops.op_builder.FusedAdamBuilder,
|
# echo 'export PATH=/opt/conda/bin:$PATH' >> ~/.bashrc && \
|
||||||
# deepspeed.ops.op_builder.CPUAdamBuilder,
|
# echo "source activate ${CONDA_ENV_NAME}" > ~/.bashrc
|
||||||
# deepspeed.ops.op_builder.CPUAdagradBuilder,
|
|
||||||
# deepspeed.ops.op_builder.CPULionBuilder,
|
|
||||||
# deepspeed.ops.op_builder.EvoformerAttnBuilder,
|
|
||||||
# deepspeed.ops.op_builder.FPQuantizerBuilder,
|
|
||||||
# deepspeed.ops.op_builder.FusedLambBuilder,
|
|
||||||
# deepspeed.ops.op_builder.FusedLionBuilder,
|
|
||||||
# deepspeed.ops.op_builder.QuantizerBuilder,
|
|
||||||
# deepspeed.ops.op_builder.RaggedOpsBuilder,
|
|
||||||
# deepspeed.ops.op_builder.RandomLTDBuilder,
|
|
||||||
# deepspeed.ops.op_builder.SparseAttnBuilder,
|
|
||||||
# deepspeed.ops.op_builder.SpatialInferenceBuilder,
|
|
||||||
# deepspeed.ops.op_builder.TransformerBuilder,
|
|
||||||
# deepspeed.ops.op_builder.StochasticTransformerBuilder,
|
|
||||||
# ]
|
|
||||||
|
|
||||||
# for builder in builders:
|
|
||||||
# print(f"Compiling {builder.__name__}")
|
|
||||||
# builder().load()
|
|
||||||
|
|
||||||
# if __name__ == "__main__":
|
|
||||||
# compile_ops()
|
|
||||||
# EOF
|
|
||||||
# python compile_deepspeed_ops.py
|
|
||||||
# ds_report
|
|
||||||
# # clean up
|
|
||||||
# # rm -f deepspeed/git_version_info_installed.py
|
|
||||||
# # rm -rf dist build deepspeed.egg-info
|
|
||||||
# # python setup.py bdist_wheel
|
|
||||||
# # DS_BUILD_OPS=${DS_BUILD_OPS} pip install -v dist/deepspeed*.whl
|
|
||||||
# # DS_BUILD_OPS=${DS_BUILD_OPS} pip install -v -r requirements/requirements.txt
|
|
||||||
# # pip install numpy==1.22.4 # ImportError: cannot import name 'BUFSIZE' from 'numpy' (/opt/conda/envs/deepspeed/lib/python3.10/site-packages/numpy/__init__.py) wait for fix in numpy=2.0.0
|
|
||||||
# EOT
|
|
||||||
|
|
||||||
# # install transformers and flash-attn
|
|
||||||
# RUN <<EOT
|
|
||||||
# #!/bin/bash
|
|
||||||
# source /opt/conda/etc/profile.d/conda.sh
|
|
||||||
# conda activate ${CONDA_ENV_NAME}
|
|
||||||
# # install transformers
|
|
||||||
# git clone https://github.com/huggingface/transformers ${STAGE_DIR}/transformers
|
|
||||||
# cd ${STAGE_DIR}/transformers
|
|
||||||
# python3 ./setup.py develop
|
|
||||||
# python3 -m pip install -U --no-cache-dir "pydantic<2"
|
|
||||||
# # install flash-attn
|
|
||||||
# # pip install packaging -i https://pypi.org/simple/ --trusted-host pypi.org
|
|
||||||
# pip install flash-attn --no-build-isolation -i https://pypi.org/simple/ --trusted-host pypi.org
|
|
||||||
# EOT
|
|
||||||
|
|
||||||
# # other packages
|
|
||||||
# ENV TORCH_CUDA_ARCH_LIST="80;86;89;90"
|
|
||||||
# RUN <<EOT
|
|
||||||
# #!/bin/bash
|
|
||||||
# source /opt/conda/etc/profile.d/conda.sh
|
|
||||||
# conda activate ${CONDA_ENV_NAME}
|
|
||||||
# pip3 install optimum
|
|
||||||
# pip3 install peft tiktoken \
|
|
||||||
# tqdm matplotlib seaborn numpy pandas scikit-learn diffusers \
|
|
||||||
# huggingface_hub spacy blobfile pycocotools \
|
|
||||||
# open_clip_torch \
|
|
||||||
# zstandard -i https://pypi.org/simple/ --trusted-host pypi.org
|
|
||||||
# EOT
|
|
||||||
|
|
||||||
# ARG DEEPSPEED_TRAIN='/data/train_data'
|
|
||||||
# ENV DEEPSPEED_TRAIN=DEEPSPEED_TRAIN
|
|
||||||
# ARG DEEPSPEED_VALIDATION='/data/validation_data'
|
|
||||||
# ENV DEEPSPEED_VALIDATION=DEEPSPEED_VALIDATION
|
|
||||||
# ARG NCCL_SOCKET_IFNAME='eth0'
|
|
||||||
|
|
||||||
# # RUN echo 'export CUDA_HOME=/usr/local/cuda' >> ~/.bashrc && \
|
|
||||||
# # echo 'export PATH=${CUDA_HOME}/bin:${PATH}' >> ~/.bashrc && \
|
|
||||||
# # echo 'export CUTLASS_PATH=/opt/cutlass' >> ~/.bashrc && \
|
|
||||||
# # echo 'export PATH=/opt/conda/bin:$PATH' >> ~/.bashrc && \
|
|
||||||
# # echo "source activate ${CONDA_ENV_NAME}" > ~/.bashrc
|
|
||||||
|
|
||||||
CMD ["/usr/sbin/sshd", "-D"]
|
CMD ["/usr/sbin/sshd", "-D"]
|
||||||
# CMD ["/bin/bash", "-c", "/usr/sbin/sshd -D & while true; do sleep 1000; done"]
|
# CMD ["/bin/bash", "-c", "/usr/sbin/sshd -D & while true; do sleep 1000; done"]
|
||||||
|
|||||||
Reference in New Issue
Block a user