You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
1, how to reproduce:
mkdir /path_to_repo/PaddleCustomDevice/backends/intel_hpu/build
cd /path_to_repo/PaddleCustomDevice/backends/intel_hpu/build
cmake ..
make -j16
pip install --force-reinstall dist/*.whl
cd ../custom_ops
python setup.py install
cd /path_to_repo/PaddleCustomDevice/backends/intel_hpu/tests/unittests/
python test_reduce_sum.py
python test_gather_nd.py
2, when using commit c35130f, there is no issue;
but when using ad9b40e, there will have
Traceback (most recent call last):
File "/workspace/pdpd_automation/repo/PaddleCustomDevice/backends/intel_hpu/tests/unittests/test_gather.py", line 57, in test_check_output
self.check_output_with_place(self.place)
File "/workspace/pdpd_automation/repo/PaddleCustomDevice/python/tests/op_test.py", line 2237, in check_output_with_place
set_flags({"FLAGS_use_onednn": False})
File "/usr/local/lib/python3.10/dist-packages/paddle/base/framework.py", line 153, in set_flags
raise ValueError(
ValueError: Flag FLAGS_use_onednn cannot set its value through this function.
commit ad9b40e (HEAD -> develop)
Author: Yuqiang Ge [email protected]
Date: Mon Aug 11 10:25:26 2025 +0800
update paddle and fix sdaa test error (#1876)
3, this regression caused about 50% INTEL_HPU Unittest cases failed