Skip to content

E2E Nightly_OnDemand Tests #294

E2E Nightly_OnDemand Tests

E2E Nightly_OnDemand Tests #294

name: E2E Nightly_OnDemand Tests
on:
schedule:
# GMT+8 21:00 every day
- cron: '0 13 * * *'
workflow_dispatch:
inputs:
python:
required: false
type: string
default: '3.10'
description: Specify python version
triton:
required: false
type: string
default: ''
description: Specify triton commit, use pytorch pined commit by default
suite:
required: true
type: string
default: 'huggingface'
description: Dynamo benchmarks test suite. huggingface,timm_models,torchbench. Delimiter is comma
dt:
required: true
type: string
default: 'float32'
description: Data precision of the test.float32,bfloat16,float16,amp_bf16,amp_fp16. Delimiter is comma
mode:
required: true
type: string
default: 'inference'
description: inference,training. Delimiter is comma
scenario:
required: true
type: string
default: 'accuracy'
description: accuracy,performance. Delimiter is comma
model:
required: false
type: string
default: ''
description: If set, will only launch this one
permissions: read-all
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}-${{ inputs.suite }}-${{ inputs.dt }}-${{ inputs.mode }}-${{ inputs.scenario }}-${{ inputs.triton }}-${{ inputs.model }}
cancel-in-progress: true
jobs:
Inductor-XPU-E2E-Nightly-Tests:
runs-on: pvc_e2e
# Don't run on forked repos
if: github.repository_owner == 'intel'
timeout-minutes: 900
outputs:
TORCH_BRANCH_ID: ${{ steps.pinned.outputs.TORCH_BRANCH_ID }}
TORCH_COMMIT_ID: ${{ steps.pinned.outputs.TORCH_COMMIT_ID }}
DRIVER_VERSION: ${{ steps.pinned.outputs.DRIVER_VERSION }}
BUNDLE_VERSION: ${{ steps.pinned.outputs.BUNDLE_VERSION }}
OS_PRETTY_NAME: ${{ steps.pinned.outputs.OS_PRETTY_NAME }}
GCC_VERSION: ${{ steps.pinned.outputs.GCC_VERSION }}
TORCHBENCH_COMMIT_ID: ${{ steps.pinned.outputs.TORCHBENCH_COMMIT_ID }}
TORCHVISION_COMMIT_ID: ${{ steps.pinned.outputs.TORCHVISION_COMMIT_ID }}
TORCHAUDIO_COMMIT_ID: ${{ steps.pinned.outputs.TORCHAUDIO_COMMIT_ID }}
# TORCHTEXT_COMMIT_ID: ${{ steps.pinned.outputs.TORCHTEXT_COMMIT_ID }}
TRANSFORMERS_VERSION: ${{ steps.pinned.outputs.TRANSFORMERS_VERSION }}
TIMM_COMMIT_ID: ${{ steps.pinned.outputs.TIMM_COMMIT_ID }}
TRITON_COMMIT_ID: ${{ steps.pinned.outputs.TRITON_COMMIT_ID }}
steps:
- name: Checkout torch-xpu-ops
uses: actions/checkout@v4
- name: Prepare Conda ENV
run: |
which conda && conda clean -ay
conda remove --all -y -n e2e_ci || rm -rf $(dirname ${CONDA_EXE})/../envs/e2e_ci
conda create -n e2e_ci python=${{ inputs.python }} cmake ninja -y
source activate e2e_ci
conda install -c intel mkl-static mkl-include -y
pip install pandas scipy tqdm
- name: Prepare Stock Pytorch
run: |
pwd
cd ../ && rm -rf pytorch
source activate e2e_ci
git clone -b main https://github.com/pytorch/pytorch pytorch
cd pytorch
# apply PRs for stock pytorch
pip install requests
python ../torch-xpu-ops/.github/scripts/apply_torch_pr.py
git status && git show -s
git submodule sync && git submodule update --init --recursive
rm -rf third_party/torch-xpu-ops && cp -r ../torch-xpu-ops third_party/
# Workaround for torch-xpu-ops ci test
sed -i "s/checkout --quiet \${TORCH_XPU_OPS_COMMIT}/log -n 1/g" caffe2/CMakeLists.txt
- name: Identify pinned versions
id: pinned
run: |
cd ../pytorch
if [ -z ${{ inputs.triton }} ]; then
echo "TRITON_COMMIT_ID=$(<.ci/docker/ci_commit_pins/triton-xpu.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
else
echo "TRITON_COMMIT_ID=${{ inputs.triton }}" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
fi
echo "TORCH_BRANCH_ID=$(git rev-parse --abbrev-ref HEAD)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
echo "TORCH_COMMIT_ID=$(git rev-parse HEAD)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
echo "TORCHBENCH_COMMIT_ID=$(<third_party/torch-xpu-ops/.github/ci_commit_pins/torchbench.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
echo "TORCHVISION_COMMIT_ID=$(<.github/ci_commit_pins/vision.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
echo "TORCHAUDIO_COMMIT_ID=$(<.github/ci_commit_pins/audio.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
# echo "TORCHTEXT_COMMIT_ID=$(<.github/ci_commit_pins/text.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
echo "TRANSFORMERS_VERSION=$(<.ci/docker/ci_commit_pins/huggingface.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
echo "TIMM_COMMIT_ID=$(<.ci/docker/ci_commit_pins/timm.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
echo "MODEL_ONLY_NAME=${{ inputs.model }}" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
source /opt/intel/oneapi/compiler/latest/env/vars.sh
echo "DRIVER_VERSION=$(dkms status 2>&1 |grep 'intel-i915-dkms' |sed 's/.*\///;s/,.*//')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
echo "BUNDLE_VERSION=$(dpcpp --version 2>&1 |grep 'DPC++/C++' |sed 's/.*(//;s/).*//')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
. /etc/os-release
echo "OS_PRETTY_NAME=${PRETTY_NAME}" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
echo "GCC_VERSION=$(gcc -dumpversion)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
echo ${GITHUB_ENV}
- name: Triton Installation
run: |
source activate e2e_ci
cd ../pytorch
TRITON_REPO="https://github.com/intel/intel-xpu-backend-for-triton"
echo ${TRITON_REPO}@${TRITON_COMMIT_ID}
pip install --force-reinstall "git+${TRITON_REPO}@${TRITON_COMMIT_ID}#subdirectory=python"
- name: Build Pytorch XPU
run: |
source activate e2e_ci
cd ../pytorch
pip install -r requirements.txt
export USE_XPU=1
source /opt/intel/oneapi/compiler/latest/env/vars.sh
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
python setup.py bdist_wheel
pip install --force-reinstall dist/*.whl
- name: Show GITHUB_ENV
run: |
echo "$GITHUB_ENV"
rm -rf ../pytorch/inductor_log
rm -rf /tmp/torchinductor_*
- name: Nightly Huggingface FP32 Inference Accuracy Test
if: ${{ !inputs.suite }}
uses: ./.github/actions/inductor-xpu-e2e-test
with:
suite: huggingface
env_prepare: true
dt: float32
mode: inference
scenario: accuracy
hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
- name: Nightly Huggingface BF16 Inference Accuracy Test
if: ${{ !inputs.suite }}
uses: ./.github/actions/inductor-xpu-e2e-test
with:
suite: huggingface
dt: bfloat16
mode: inference
scenario: accuracy
hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
- name: Nightly Huggingface FP16 Inference Accuracy Test
if: ${{ !inputs.suite }}
uses: ./.github/actions/inductor-xpu-e2e-test
with:
suite: huggingface
dt: float16
mode: inference
scenario: accuracy
hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
- name: Nightly Huggingface FP32 Training Accuracy Test
if: ${{ !inputs.suite }}
uses: ./.github/actions/inductor-xpu-e2e-test
with:
suite: huggingface
dt: float32
mode: training
scenario: accuracy
hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
- name: Nightly Huggingface BF16 Training Accuracy Test
if: ${{ !inputs.suite }}
uses: ./.github/actions/inductor-xpu-e2e-test
with:
suite: huggingface
dt: bfloat16
mode: training
scenario: accuracy
hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
- name: Nightly Huggingface FP16 Training Accuracy Test
if: ${{ !inputs.suite }}
uses: ./.github/actions/inductor-xpu-e2e-test
with:
suite: huggingface
dt: float16
mode: training
scenario: accuracy
hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
- name: Nightly Torchbench BF16 Training Accuracy Test
if: ${{ !inputs.suite }}
uses: ./.github/actions/inductor-xpu-e2e-test
with:
suite: torchbench
dt: bfloat16
mode: training
scenario: accuracy
env_prepare: true
hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
- name: Nightly Timm_models FP16 Training Accuracy Test
if: ${{ !inputs.suite }}
uses: ./.github/actions/inductor-xpu-e2e-test
with:
suite: timm_models
dt: float16
mode: training
scenario: accuracy
env_prepare: true
hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
- name: OnDemand Test (${{ inputs.suite }} ${{ inputs.dt }} ${{ inputs.mode }} ${{ inputs.scenario }})
if: ${{ inputs.suite }}
uses: ./.github/actions/inductor-xpu-e2e-test
with:
suite: ${{ inputs.suite }}
env_prepare: true
dt: ${{ inputs.dt }}
mode: ${{ inputs.mode }}
scenario: ${{ inputs.scenario }}
hf_token: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
- name: Summarize archieve files
if: always()
run: |
rm -rf ${{ github.workspace }}/upload_files
cp -r ${{ github.workspace }}/../pytorch/inductor_log ${{ github.workspace }}/upload_files
failed_case=$(grep "Real failed: models: *[1-9]" ${{ github.workspace }}/upload_files/summary_accuracy.log |wc -l || true)
if [ ${failed_case} -ne 0 ];then
grep -E "Real failed: models: [1-9]|Summary for" ${{ github.workspace }}/summary_accuracy.log
exit 1
fi
- name: Upload Inductor XPU E2E Data
if: always()
uses: actions/upload-artifact@v4
with:
name: Inductor-XPU-E2E-Data-${{ github.event.pull_request.number || github.sha }}
path: ${{ github.workspace }}/upload_files
Tests-Failure-And-Report:
if: always()
runs-on: pvc_e2e
permissions:
issues: write
env:
GH_TOKEN: ${{ github.token }}
needs: Inductor-XPU-E2E-Nightly-Tests
steps:
- name: Report github issue for XPU OPS nightly
if: github.repository_owner == 'intel'
run: |
set -xe
# Test env
build_url="${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
repo="${{ github.repository }}"
TORCH_BRANCH_ID="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TORCH_BRANCH_ID }}"
TORCH_COMMIT_ID="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TORCH_COMMIT_ID }}"
DRIVER_VERSION="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.DRIVER_VERSION }}"
BUNDLE_VERSION="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.BUNDLE_VERSION }}"
OS_PRETTY_NAME="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.OS_PRETTY_NAME }}"
GCC_VERSION="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.GCC_VERSION }}"
TORCHBENCH_COMMIT_ID="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TORCHBENCH_COMMIT_ID }}"
TORCHVISION_COMMIT_ID="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TORCHVISION_COMMIT_ID }}"
TORCHAUDIO_COMMIT_ID="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TORCHAUDIO_COMMIT_ID }}"
# TORCHTEXT_COMMIT_ID="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TORCHTEXT_COMMIT_ID }}"
TRANSFORMERS_VERSION="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TRANSFORMERS_VERSION }}"
TIMM_COMMIT_ID="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TIMM_COMMIT_ID }}"
TRITON_COMMIT_ID="${{ needs.Inductor-XPU-E2E-Nightly-Tests.outputs.TRITON_COMMIT_ID }}"
# Test status
if [ "${{ needs.Inductor-XPU-E2E-Nightly-Tests.result }}" == "success" ];then
test_status=Success
elif [ "${{ needs.Inductor-XPU-E2E-Nightly-Tests.result }}" == "failure" ];then
test_status=Failure
cc_comment="CC ${{ secrets.NIGHTLY_EMAIL_LIST }}"
else
test_status=None
exit 0
fi
# Test Type
if [ "${GITHUB_EVENT_NAME}" == "workflow_dispatch" ];then
test_type="On-demand"
test_issue_id=426
cc_comment="CC @${GITHUB_TRIGGERING_ACTOR}"
else
test_type="Nightly"
test_issue_id=432
fi
# Test report
echo -e "$cc_comment\n**${test_status}** $test_type Test on $(date +'%F'), See: $build_url\n" > ${{ github.workspace }}/report.txt
printf "Torch-xpu-ops | PyTorch | Triton\n--- | --- | ---\n${GITHUB_WORKFLOW_SHA:0:7} on ${GITHUB_REF_NAME} | " >> ${{ github.workspace }}/report.txt
printf "[${TORCH_COMMIT_ID:0:7}](https://github.com/pytorch/pytorch/commit/${TORCH_COMMIT_ID:0:7}) on $TORCH_BRANCH_ID | " >> ${{ github.workspace }}/report.txt
echo -e "[${TRITON_COMMIT_ID:0:7}](https://github.com/intel/intel-xpu-backend-for-triton/commit/${TRITON_COMMIT_ID:0:7}) \n" >> ${{ github.workspace }}/report.txt
printf "Transformers | Timm | Torchbench | Torchvision | Torchaudio\n--- | --- | --- | --- | ---\n" >> ${{ github.workspace }}/report.txt
printf "[${TRANSFORMERS_VERSION:0:7}](https://github.com/huggingface/transformers/commit/${TRANSFORMERS_VERSION:0:7}) | " >> ${{ github.workspace }}/report.txt
printf "[${TIMM_COMMIT_ID:0:7}](https://github.com/huggingface/pytorch-image-models/commit/${TIMM_COMMIT_ID:0:7}) | " >> ${{ github.workspace }}/report.txt
printf "[${TORCHBENCH_COMMIT_ID:0:7}](https://github.com/pytorch/benchmark/commit/${TORCHBENCH_COMMIT_ID:0:7}) | " >> ${{ github.workspace }}/report.txt
printf "[${TORCHVISION_COMMIT_ID:0:7}](https://github.com/pytorch/vision/commit/${TORCHVISION_COMMIT_ID:0:7}) | " >> ${{ github.workspace }}/report.txt
echo -e "[${TORCHAUDIO_COMMIT_ID:0:7}](https://github.com/pytorch/audio/commit/${TORCHAUDIO_COMMIT_ID:0:7}) \n" >> ${{ github.workspace }}/report.txt
printf "Device | OS | GCC | Python | Driver(DKMS) | Bundle(DPCPP)\n--- | --- | --- | --- | --- | ---\n" >> ${{ github.workspace }}/report.txt
echo -e "$RUNNER_NAME | $OS_PRETTY_NAME | $GCC_VERSION | ${{ inputs.python }} | $DRIVER_VERSION| $BUNDLE_VERSION \n" >> ${{ github.workspace }}/report.txt
if [ "${GITHUB_EVENT_NAME}" == "workflow_dispatch" ];then
test_scope="${{ inputs.suite }}/${{ inputs.dt }}/${{ inputs.mode }}/${{ inputs.scenario }}"
if [ "${{ inputs.triton }}" != "" ];then
test_scope+="; triton=${{ inputs.triton }}"
fi
if [ "${{ inputs.model }}" != "" ];then
test_scope+="; model=${{ inputs.model }}"
fi
echo -e "Inputs | $test_scope\n--- | --- \n" >> ${{ github.workspace }}/report.txt
fi
# Report
report_txt=$(cat ${{ github.workspace }}/report.txt)
gh --repo $repo issue comment $test_issue_id --body "$report_txt"