Migrate to 1ES pipeline (#95)

.github/workflows/examples.yml

@@ -10,22 +10,19 @@ on:
 
 jobs:
   examples:
-    runs-on: [self-hosted, linux, gpu]
+    runs-on: [self-hosted, 1ES.Pool=agl-runner-gpu]
     timeout-minutes: 60
     strategy:
       matrix:
         setup: [stable, latest]
       fail-fast: false
-    container:
-      image: ghcr.io/microsoft/agent-lightning/base:latest
-      options: --gpus all --ipc=host --interactive --tty
     steps:
       - name: Check GPU status
         run: nvidia-smi
       - uses: actions/checkout@v4
       - name: Create a virtual environment
         run: python3 -m venv .venv
-      - name: Install deps inside the container (${{ matrix.setup }})
+      - name: Install dependencies (${{ matrix.setup }})
         run: |
           . .venv/bin/activate
           ./scripts/setup_${{ matrix.setup }}_gpu.sh

scripts/build_vm_image.sh

@@ -0,0 +1,63 @@
+#!/bin/bash
+
+# Based on: Standard_NC24ads_A100_v4
+# With: Canonical:ubuntu-24_04-lts:server:latest
+# Secure boot is off.
+
+# This script is not designed to be run automatically.
+
+set -ex
+
+# Build essentials are required.
+sudo apt-get clean
+sudo apt-get update
+sudo apt-get install -y \
+    git \
+    wget \
+    curl \
+    build-essential \
+    software-properties-common \
+    python3-dev \
+    python3-pip \
+    python3-venv \
+    graphviz \
+    unzip \
+    tmux \
+    vim \
+    git-lfs \
+    nodejs
+
+git lfs install
+
+# VM with GPU needs to install drivers. Reference:
+# https://docs.microsoft.com/en-us/azure/virtual-machines/linux/n-series-driver-setup
+sudo apt update && sudo apt install -y ubuntu-drivers-common
+sudo ubuntu-drivers install
+sudo reboot now
+
+# Install CUDA Toolkit
+wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/cuda-keyring_1.1-1_all.deb
+sudo dpkg -i cuda-keyring_1.1-1_all.deb
+sudo apt-get update
+sudo apt-get -y install cuda-toolkit-12-8
+sudo reboot now
+
+# Add paths globally
+sudo bash -c "cat > /etc/profile.d/cuda.sh" <<'EOF'
+export PATH=/usr/local/cuda/bin:$PATH
+export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
+EOF
+sudo chmod +x /etc/profile.d/cuda.sh
+
+# Disable the periodical apt-get upgrade.
+# Sometimes, unattended upgrade blocks apt-get install
+sudo sed -i -e "s/Update-Package-Lists \"1\"/Update-Package-Lists \"0\"/g" /etc/apt/apt.conf.d/10periodic
+sudo sed -i -e "s/Update-Package-Lists \"1\"/Update-Package-Lists \"0\"/g" /etc/apt/apt.conf.d/20auto-upgrades
+sudo sed -i -e "s/Unattended-Upgrade \"1\"/Unattended-Upgrade \"0\"/g" /etc/apt/apt.conf.d/20auto-upgrades
+sudo systemctl stop apt-daily.timer apt-daily-upgrade.timer
+sudo systemctl stop apt-daily.service apt-daily-upgrade.service
+sudo systemctl mask apt-daily.timer apt-daily-upgrade.timer
+sudo systemctl mask apt-daily.service apt-daily-upgrade.service
+
+# Deprovision and prepare for generalized image
+sudo waagent -deprovision+user

scripts/setup_latest_gpu.sh

@@ -4,10 +4,10 @@ python -m pip install --upgrade --no-cache-dir pip
 
 pip install --no-cache-dir packaging ninja numpy pandas ipython ipykernel gdown wheel setuptools
 # This has to be pinned for VLLM to work.
-pip install --no-cache-dir torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu128
+pip install --no-cache-dir torch==2.8.0 torchvision==0.23.0 --index-url https://download.pytorch.org/whl/cu128
 pip install --no-cache-dir flash-attn --no-build-isolation
 # This must match pytorch version.
-pip install --no-cache-dir vllm==0.10.1.1
+pip install --no-cache-dir vllm==0.10.2
 # Latest VERL release version.
 pip install --no-cache-dir verl