Add a BERT example using Intel Neural Compressor SmoothQuant #411
Conversation
|
do we need enable it in CI pipeline? |
Yes, we do need to enable it in the CI pipeline. Could you please provide some guidance or suggestions on how to do this? It would be of great help. |
Signed-off-by: yuwenzho <[email protected]>
Signed-off-by: yuwenzho <[email protected]>
yuwenzho
force-pushed
the
yuwenzho/inc_sq_example
branch
from
d9d7726 to
87fead6
Compare
|
I enabled |
examples/test/test_bert_ptq_cpu.py
Outdated
| @@ -23,12 +23,15 @@ def setup(): | |||
| @pytest.mark.parametrize("search_algorithm", ["tpe"]) | |||
| @pytest.mark.parametrize("execution_order", ["joint"]) | |||
| @pytest.mark.parametrize("system", ["local_system", "aml_system", "docker_system"]) | |||
| @pytest.mark.parametrize("olive_json", ["bert_ptq_cpu.json"]) | |||
| @pytest.mark.parametrize("olive_json", ["bert_ptq_cpu.json", "bert_inc_ptq_cpu.json", "bert_inc_smoothquant_ptq_cpu.json"]) | |||
There was a problem hiding this comment.
Emm, if so, current Bert pipeline will run with about 3x times than before(about 1h). As it is only test the inc quantizaion passes, maybe we do not need to run the complete E2E tests?
I believe INC team already had their tests but not in Olive just like other ONNX passes.
There was a problem hiding this comment.
Thanks for your suggestion. I removed Bert INC test from the CI pipeline. All INC-related examples are tested internally by INC team.
yuwenzho
force-pushed
the
yuwenzho/inc_sq_example
branch
from
87fead6 to
1619952
Compare
|
/azp run |
|
No commit pushedDate could be found for PR 411 in repo microsoft/Olive |
|
/azp run |
|
Azure Pipelines successfully started running 2 pipeline(s). |
Describe your changes
Add a BERT example using Intel® Neural Compressor SmoothQuant.
Add an introduction of Intel® Neural Compressor SmoothQuant to the README.
Checklist before requesting a review
pre-commit run --all-files(Optional) Issue link