Skip to content

Conda Test

Conda Test #251

name: Conda build and test
on:
workflow_dispatch:
schedule:
- cron: "5 2 * * *"
permissions: read-all
env:
BACKEND: XPU
TRITON_DISABLE_LINE_INFO: 1
GH_TOKEN: ${{ github.token }}
jobs:
integration-tests:
name: Integration tests
runs-on:
- glados
- spr
- runner-0.0.13
strategy:
matrix:
python:
- "3.9"
- "3.10"
defaults:
run:
shell: bash -noprofile --norc -eo pipefail -c "source /home/runner/intel/oneapi/setvars.sh > /dev/null; source {0}"
steps:
- name: Checkout repository
uses: actions/checkout@v4
- name: Calculate env
run: |
echo $HOME/miniconda3/bin >>$GITHUB_PATH
- name: Load conda cache
id: conda-cache
uses: ./.github/actions/load
env:
CACHE_NUMBER: 4
with:
path: $HOME/miniconda3/envs/triton
key: conda-py${{ matrix.python }}-${{ hashFiles('scripts/triton.yml', 'python/pyproject.toml', 'python/setup.py', '.github/pins/ipex.txt', '.github/pins/pytorch.txt') }}-${{ env.CACHE_NUMBER }}
- name: Update conda env
if: ${{ steps.conda-cache.outputs.status == 'miss' }}
run: |
conda create -n triton --override-channels -c conda-forge python=${{ matrix.python }}.*
conda env update -f scripts/triton.yml
find /home/runner/intel/oneapi/ \( -name '*.so' -or -name '*.so.*' \) -exec cp -n {} $HOME/miniconda3/envs/triton/lib \;
ln -snf /usr/include/level_zero $HOME/miniconda3/envs/triton/bin/../x86_64-conda-linux-gnu/sysroot/usr/include/level_zero
find /usr -name libze_\* -exec cp -n {} $HOME/miniconda3/envs/triton/lib \;
- name: Add conda info to log
run: |
conda info
conda list -n triton
- name: Install latest nightly wheels
uses: ./.github/actions/install-wheels
with:
gh_token: ${{ secrets.GITHUB_TOKEN }}
install_cmd: conda run --no-capture-output -n triton pip install
python_version: ${{ matrix.python }}
wheels_pattern: '{intel_extension_for_pytorch-*,torch-*}'
- name: Build Triton
run: |
set -x
export DEBUG=1
cd python
conda run --no-capture-output -n triton pip install --no-build-isolation -e '.[build,tests,tutorials]'
- name: Run core tests
env:
# FIXME https://github.com/intel/intel-xpu-backend-for-triton/issues/806
# FIXME https://github.com/intel/intel-xpu-backend-for-triton/issues/841
TRITON_TEST_SKIPLIST_DIR: scripts/skiplist/conda
run: |
set -x
conda run --no-capture-output -n triton bash -v -x scripts/test-triton.sh
- name: Run E2E test
run: |
cd ../pytorch || {
PYTORCH_COMMIT_ID=$(<.github/pins/pytorch.txt)
cd ..
git clone --single-branch -b dev/triton-test-3.0 --recurse-submodules https://github.com/Stonepia/pytorch.git
cd pytorch
git branch pin-branch $PYTORCH_COMMIT_ID
git switch pin-branch
}
TRANSFORMERS_VERSION="$(<.ci/docker/ci_commit_pins/huggingface.txt)"
conda run -n triton pip install pyyaml pandas scipy numpy psutil pyre_extensions torchrec transformers==$TRANSFORMERS_VERSION
# Set WORKSPACE for inductor_xpu_test.sh to make sure it creates "inductor_log" outside of pytorch cloned directory
export WORKSPACE=$GITHUB_WORKSPACE
# TODO: Find the fastest Hugging Face model
conda run --no-capture-output -n triton $GITHUB_WORKSPACE/scripts/inductor_xpu_test.sh huggingface float32 inference accuracy xpu 0 static 1 0 AlbertForMaskedLM
# The script above always returns 0, so we need an additional check to see if the accuracy test passed
cat $WORKSPACE/inductor_log/*/*/*.csv
grep AlbertForMaskedLM $WORKSPACE/inductor_log/*/*/*.csv | grep -q ,pass,
- name: Save conda cache
if: ${{ steps.conda-cache.outputs.status == 'miss' }}
uses: ./.github/actions/save
with:
path: ${{ steps.conda-cache.outputs.path }}
dest: ${{ steps.conda-cache.outputs.dest }}