reproduce steps
$ https://github.com/yao-matrix/transformers.git
$ cd ./transformers
$ git checkout remove-ipex
$ tests/quantization/autoround/test_auto_round.py::AutoRoundTest::test_convert_from_awq_cpu
results
case failed.
context
we are removing ipex from Hugging Face transformers per plan, this is the only left case we cannot pass, and transformers is deprecating autoawq w/ gptqmodel for awq models, could we update this case to go through a non-ipex backend, like through gptqmodel's torch_fused backend or kernels backend which has been enabled by @jiqing-feng.