ci: add test for Huggingface Accelerate #1257
Conversation
dvrogozh
force-pushed
the
accelerate
branch
15 times, most recently
from
b728584 to
6666291
Compare
| - name: Prepare Stock XPU Pytorch | ||
| run: | | ||
| source activate $CONDA_ENV_NAME | ||
| if [ -z "${{ inputs.nightly_whl }}" ]; then |
There was a problem hiding this comment.
Do we plan this test only target for torch cpu nightly wheel? latest nightly or a specific nightly version?
There was a problem hiding this comment.
I suggest to target it for latest nightly triggered per schedule daily. @chuanqi129 : are you ok if I will drop input parameter to set nightly version?
There was a problem hiding this comment.
I dropped inputs.nightly_whl for now and just install latest nightly. Note that this parameter in any case would need a change since it controls only torch version and does not allow to set version for torchvision and torchaudio. So, if needed, we can add these options later on.
| ZE_AFFINITY_MASK: 0 | ||
| PARSE_JUNIT: ${{ github.workspace }}/torch-xpu-ops/.github/scripts/parse-junitxml.py | ||
| steps: | ||
| - name: Checkout torch-xpu-ops |
There was a problem hiding this comment.
Seems we don't need checkout torch-xpu-ops
There was a problem hiding this comment.
We need it. Workflow uses:
- .github/scripts/parse-junitxml.py
- .github/actions/print-environment/action.yml
from torch-xpu-ops repo. However, I forgot to add .py script to the dependencies of this workflow and I also I forgot to drop spec.py which is Transformers specific - I will fix that.
There was a problem hiding this comment.
However, I forgot to add .py script to the dependencies of this workflow and I also I forgot to drop spec.py which is Transformers specific - I will fix that.
Done.
For Accelerate it's not actually expansion to ecosystem. It's closing a test gap for Transformers. Note - Accelerate is a prerequisite library for Transformers. If something breaks in Accelerate there is high chance that Transformer tests will also start to fall apart. |
dvrogozh
force-pushed
the
accelerate
branch
2 times, most recently
from
f3a9804 to
c959881
Compare
Script parses output of JUnit result XML files which can be generated by pytest -junit-xml=report.xml option. The intended usage of the script: * Print report tables in Markdown format. Use script arguments to control output: * `parse-junitxml.py --stats` to print stats table * `parse-junitxml.py --failed` to print table with failed cases * `parse-junitxml.py --skipped` to print table with skipped cases * etc. * Print report tables in Json or Json-line formats. Add `--json` to the script command line. Signed-off-by: Dmitry Rogozhkin <[email protected]>
Signed-off-by: Dmitry Rogozhkin <[email protected]>
Signed-off-by: Dmitry Rogozhkin <[email protected]>
Changes:
--junitxml=option