Merge pull request #17 from foundation-model-stack/test-gpl

Create new image using clang

.github/workflows/build-and-push-nightly-pytorch-compile-clang.yml

@@ -0,0 +1,26 @@
+name: build-and-push-nightly-pytorch-compile-clang
+
+on:
+  schedule:
+    # Runs "nightly, at 3am" (see https://crontab.guru)
+    - cron: '0 3 * * *'
+  workflow_dispatch:
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    environment: image-build-and-push
+    steps:
+      # docker login source https://github.com/docker/login-action
+      - uses: docker/login-action@v2
+        with:
+          registry: ghcr.io
+          # Note: these values can be configured under https://github.com/foundation-model-stack/base-images/settings/environments/614355731/edit
+          username: ${{ secrets.GH_REGISTRY_USER }}
+          password: ${{ secrets.GH_REGISTRY_PUSH_TOKEN }}
+
+      # Checks-out your repository under $GITHUB_WORKSPACE, so your job can access it
+      - uses: actions/checkout@v3
+
+      - name: build and push
+        run: ./nightly-pytorch-compile-clang/buildAndPush.sh

nightly-pytorch-compile-clang/Dockerfile

@@ -0,0 +1,70 @@
+# switch to miniconda3 base as pytorch base uses python3.7
+FROM continuumio/miniconda3:latest
+
+ENV HOME=/homedir \
+    AIM_UI_TELEMETRY_ENABLED=0
+
+# tools
+RUN apt-get -y -o Acquire::Max-FutureTime=86400 update && \
+    apt-get install -y gnupg vim nmon clang software-properties-common && \
+    apt-key del 7fa2af80 && \
+    add-apt-repository contrib && \
+    wget https://developer.download.nvidia.com/compute/cuda/repos/debian11/x86_64/cuda-keyring_1.0-1_all.deb && \
+    dpkg -i cuda-keyring_1.0-1_all.deb && \
+    apt-get -y -o Acquire::Max-FutureTime=86400 update && \
+    apt-get install -y cuda-compiler-12-1 cuda-nsight-systems-12-1 && \
+    apt-get -y clean
+
+RUN chgrp -R 0 /opt/conda && \
+    chmod -R a+w /opt/conda && \
+    chmod -R g=u /opt/conda
+
+# This is important for certain pytorch deps
+RUN pip install --upgrade setuptools && \
+    chmod -R g+w /opt/conda && \
+    chmod -R g=u /opt/conda
+
+# Pytorch nightly install
+RUN pip install --pre --upgrade torch torchvision --extra-index-url https://download.pytorch.org/whl/nightly/cu118 --no-cache-dir && \
+    chmod -R g+w /opt/conda && \
+    chmod -R g=u /opt/conda
+
+# permanent dependencies, we put them after torch so we don't end up installing torch twice and wasting a lot of space
+# (each install of torch is about 2GB!!!)
+# decision has been made to have them as the latest version always, as this is a nightly image after all
+RUN pip install \
+    transformers \
+    aim \
+    datasets \
+    accelerate \
+    optimum \
+    tqdm \
+    protobuf \
+    colorama \
+    scikit-learn \
+    ninja \
+    --no-cache-dir && \
+    chmod -R g+w /opt/conda && \
+    chmod -R g=u /opt/conda
+
+# xformers install (built from source)
+# deactivated until we get better workers as there isn't enough ram, cpu, or disk space to compile them anymore
+# ENV CUDA_HOME=/usr/local/cuda
+# ENV MAX_JOBS=2
+# ENV TORCH_CUDA_ARCH_LIST="7.0;7.5;8.0;8.6"
+# RUN apt-get install -y libcublas-11-8 libcublas-dev-11-8 libcusparse-11-8 libcusparse-dev-11-8 libcusolver-11-8 libcusolver-dev-11-8 libcurand-11-8 libcurand-dev-11-8 && \
+#     pip install -v -U git+https://github.com/facebookresearch/xformers.git@main#egg=xformers --no-cache-dir && \
+#     chmod -R g+w /opt/conda && \
+#     chmod -R g=u /opt/conda && \
+#     apt-get remove -y libcublas-11-8 libcublas-dev-11-8 libcusparse-11-8 libcusparse-dev-11-8 libcusolver-11-8 libcusolver-dev-11-8 libcurand-11-8 libcurand-dev-11-8 && \
+#     apt-get -y clean
+
+# setup .aim_profile (needed to get aim running)
+RUN mkdir -p ${HOME} && \
+    touch ~/.aim_profile && \
+    chmod g+w ~/.aim_profile
+
+WORKDIR /workspace
+
+RUN chgrp -R 0 . && \
+    chmod -R g=u .

nightly-pytorch-compile-clang/buildAndPush.sh

@@ -0,0 +1,23 @@
+#!/usr/bin/env bash
+
+# override image name with optional command line argument ./buildAndPush.sh <your_custom_tag>
+# otherwise it will be tagged with pytorch-latest-nightly and timestamp
+
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
+
+REGISTRY='ghcr.io'
+NAMESPACE='foundation-model-stack'
+NAME='base'
+
+cd "${SCRIPT_DIR}"
+
+TIMESTAMP=$(date "+%Y%m%d")
+
+# note: docker tags must be valid ASCII and may contain lowercase and uppercase letters, digits, underscores, periods and dashes.
+# A tag name may not start with a period or a dash and may contain a maximum of 128 characters
+VERSION="$(echo ${1:-pytorch-compile-clang-latest-nightly-$TIMESTAMP}| sed 's/[^[:alnum:]\.\_\-]//g')"
+TAG="${REGISTRY}/${NAMESPACE}/${NAME}:${VERSION}"
+
+docker build -t "${TAG}" .
+
+docker push "${TAG}"