Spaces:
Running
on
Zero
Running
on
Zero
File size: 4,384 Bytes
038856e |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 |
# Inspired by https://github.com/anibali/docker-pytorch/blob/master/dockerfiles/1.10.0-cuda11.3-ubuntu20.04/Dockerfile
# ARG COMPAT=0
ARG PERSONAL=0
# FROM nvidia/cuda:11.3.1-devel-ubuntu20.04 as base-0
FROM nvcr.io/nvidia/pytorch:22.12-py3 as base
ENV HOST docker
ENV LANG=C.UTF-8 LC_ALL=C.UTF-8
# https://serverfault.com/questions/683605/docker-container-time-timezone-will-not-reflect-changes
ENV TZ America/Los_Angeles
RUN ln -snf /usr/share/zoneinfo/$TZ /etc/localtime && echo $TZ > /etc/timezone
# git for installing dependencies
# tzdata to set time zone
# wget and unzip to download data
# [2021-09-09] TD: zsh, stow, subversion, fasd are for setting up my personal environment.
# [2021-12-07] TD: openmpi-bin for MPI (multi-node training)
RUN apt-get update && apt-get install -y --no-install-recommends \
build-essential \
cmake \
curl \
ca-certificates \
sudo \
less \
htop \
git \
tzdata \
wget \
tmux \
zip \
unzip \
zsh stow subversion fasd \
&& rm -rf /var/lib/apt/lists/*
# openmpi-bin \
# Allow running runmpi as root
# ENV OMPI_ALLOW_RUN_AS_ROOT=1 OMPI_ALLOW_RUN_AS_ROOT_CONFIRM=1
# # Create a non-root user and switch to it
# RUN adduser --disabled-password --gecos '' --shell /bin/bash user \
# && echo "user ALL=(ALL) NOPASSWD:ALL" > /etc/sudoers.d/90-user
# USER user
# All users can use /home/user as their home directory
ENV HOME=/home/user
RUN mkdir -p /home/user && chmod 777 /home/user
WORKDIR /home/user
# Set up personal environment
# FROM base-${COMPAT} as env-0
FROM base as env-0
FROM env-0 as env-1
# Use ONBUILD so that the dotfiles dir doesn't need to exist unless we're building a personal image
# https://stackoverflow.com/questions/31528384/conditional-copy-add-in-dockerfile
ONBUILD COPY dotfiles ./dotfiles
ONBUILD RUN cd ~/dotfiles && stow bash zsh tmux && sudo chsh -s /usr/bin/zsh $(whoami)
# nvcr pytorch image sets SHELL=/bin/bash
ONBUILD ENV SHELL=/bin/zsh
FROM env-${PERSONAL} as packages
# Disable pip cache: https://stackoverflow.com/questions/45594707/what-is-pips-no-cache-dir-good-for
ENV PIP_NO_CACHE_DIR=1
# # apex and pytorch-fast-transformers take a while to compile so we install them first
# TD [2022-04-28] apex is already installed. In case we need a newer commit:
# RUN pip install --upgrade --force-reinstall --global-option="--cpp_ext" --global-option="--cuda_ext" --global-option="--fast_multihead_attn" --global-option="--fmha" --global-option="--fast_layer_norm" --global-option="--xentropy" git+https://github.com/NVIDIA/apex.git#egg=apex
# xgboost conflicts with deepspeed
RUN pip uninstall -y xgboost && DS_BUILD_UTILS=1 DS_BUILD_FUSED_LAMB=1 pip install deepspeed==0.7.7
# General packages that we don't care about the version
# zstandard to extract the_pile dataset
# psutil to get the number of cpu physical cores
# twine to upload package to PyPI
RUN pip install pytest matplotlib jupyter ipython ipdb gpustat scikit-learn spacy munch einops opt_einsum fvcore gsutil cmake pykeops zstandard psutil h5py twine gdown \
&& python -m spacy download en_core_web_sm
# hydra
RUN pip install hydra-core==1.3.1 hydra-colorlog==1.2.0 hydra-optuna-sweeper==1.2.0 pyrootutils rich
# Core packages
RUN pip install transformers==4.45.2 datasets==3.0.1 pytorch-lightning==2.2.1 triton==2.3.1 wandb==0.16.3 controlnet_aux==0.0.9 timm==0.6.7 torchmetrics==1.3.2
# torchmetrics 0.11.0 broke hydra's instantiate
# For MLPerf
RUN pip install git+https://github.com/mlcommons/logging.git@2.1.0
RUN pip install accelerate==0.34.2
RUN pip install diffusers==0.30.3
RUN pip install deepspeed==0.15.2
RUN pip install sentencepiece==0.1.99
RUN pip install pillow==10.2.0
RUN pip install torch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu118
# Install FlashAttention
RUN pip install flash-attn==2.6.3
# Install CUDA extensions for fused dense
RUN pip install git+https://github.com/Dao-AILab/flash-attention@v2.6.3#subdirectory=csrc/fused_dense_lib
RUN pip install jaxtyping mediapipe gradio
RUN pip install "git+https://github.com/facebookresearch/pytorch3d.git"
RUN apt-get update && apt-get install ffmpeg libsm6 libxext6 -y
RUN pip install opencv-python==4.5.5.64
RUN pip install opencv-python-headless==4.5.5.64
RUN pip install huggingface_hub==0.24
RUN pip install numpy==1.24.4
|