Enable the a10g instance ci (#1476)

Summary:
Enable the PR CI on AWS A10G to save CI costs.

Fixes #1470

Pull Request resolved: #1476

Reviewed By: weiwangmeta

Differential Revision: D43995490

Pulled By: xuzhao9

fbshipit-source-id: 56496a247500fd141fa6f3cf1f8701a03a5d37fe

Author: xuzhao9

.github/workflows/pr-a10g.yml

@@ -0,0 +1,61 @@
+name: TorchBench PR Test on A10G
+on:
+  pull_request:
+  workflow_dispatch:
+  push:
+    branches:
+      - main
+
+env:
+  CONDA_ENV: "pr-test"
+
+jobs:
+  pr-test:
+    # AWS A10G GPU instance label: linux.g5.4xlarge.nvidia.gpu
+    # OS version: Amazon Linux 2
+    runs-on: [self-hosted, linux.g5.4xlarge.nvidia.gpu]
+    timeout-minutes: 1440 # 24 hours
+    steps:
+      - name: Checkout TorchBench
+        uses: actions/checkout@v3
+      - name: Install Conda and basic packages
+        run: |
+          bash scripts/install_basics_amzn_linux_2.sh
+      - name: Install NVIDIA Driver
+        run: |
+          bash scripts/setup_nvda_11.7.sh
+      - name: GPU Tuning
+        run: |
+          sudo nvidia-smi -pm 1
+      - name: Setup Conda Env
+        run: |
+          . ${HOME}/miniconda3/etc/profile.d/conda.sh
+          conda activate
+          python utils/python_utils.py --create-conda-env "${CONDA_ENV}"
+          conda activate "${CONDA_ENV}"
+          python utils/cuda_utils.py --install-torch-deps
+      - name: Install PyTorch nightly
+        run: |
+          . ${HOME}/miniconda3/etc/profile.d/conda.sh
+          conda activate "${CONDA_ENV}"
+          python utils/cuda_utils.py --install-torch-nightly
+      - name: Install TorchBench
+        run: |
+          . ${HOME}/miniconda3/etc/profile.d/conda.sh
+          conda activate "${CONDA_ENV}"
+          python install.py
+      - name: Validate benchmark components (Worker)
+        run: |
+          . ${HOME}/miniconda3/etc/profile.d/conda.sh
+          conda activate "${CONDA_ENV}"
+          python -m components.test.test_subprocess
+          python -m components.test.test_worker
+      - name: Validate benchmark components (Model)
+        run: |
+          . ${HOME}/miniconda3/etc/profile.d/conda.sh
+          conda activate "${CONDA_ENV}"
+          python test.py
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}-${{ github.event_name == 'workflow_dispatch' }}
+  cancel-in-progress: true

scripts/install_basics.sh

@@ -0,0 +1,13 @@
+CONDA=https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
+filename=$(basename "$CONDA")
+wget "$CONDA"
+chmod +x "$filename"
+./"$filename" -b -u
+
+sudo yum makecache --refresh
+sudo yum install -y git jq \
+                vim wget curl ninja-build cmake \
+                libglvnd-glx libsndfile
+
+. ${HOME}/miniconda3/etc/profile.d/conda.sh
+conda activate

scripts/install_basics_amzn_linux_2.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+set -ex -o pipefail
+
+# Setup NVIDIA Driver
+DRIVER_FN="NVIDIA-Linux-x86_64-515.76.run"
+wget "https://s3.amazonaws.com/ossci-linux/nvidia_driver/$DRIVER_FN"
+sudo /bin/bash "$DRIVER_FN" -s --no-drm || (sudo cat /var/log/nvidia-installer.log && false)
+nvidia-smi
+
+# Setup CUDA 11.7 and cuDNN 8.5.0.96
+wget -q https://developer.download.nvidia.com/compute/cuda/11.7.0/local_installers/cuda_11.7.0_515.43.04_linux.run -O cuda_11.7.0_515.43.04_linux.run
+sudo bash ./cuda_11.7.0_515.43.04_linux.run --toolkit --silent
+wget -q https://ossci-linux.s3.amazonaws.com/cudnn-linux-x86_64-8.5.0.96_cuda11-archive.tar.xz -O cudnn-linux-x86_64-8.5.0.96_cuda11-archive.tar.xz
+tar xJf cudnn-linux-x86_64-8.5.0.96_cuda11-archive.tar.xz
+cd cudnn-linux-x86_64-8.5.0.96_cuda11-archive && \
+sudo cp include/* /usr/local/cuda-11.7/include && \
+sudo cp lib/* /usr/local/cuda-11.7/lib64 && \
+sudo ldconfig