Khue Le

@jgong5 Thanks for your reply. I have realized that bfloat16 is faster for `batchsize=1` (tested with the first sentence in the above example). The profiling results (together with the full...

@jgong5 Yes the full output can be found in the attached file. [ipex_bfloat16_bug.txt](https://github.com/intel/intel-extension-for-pytorch/files/15455355/ipex_bfloat16_bug.txt)

@zhuhaozhe Thanks for your investigation. It's a `c3-standard-4` VM instance from Google Cloud. Please let me know if you need further information.

@WilliamTambellini Good catch! Thanks! @zhuhaozhe It's indeed 5.10. I'll upgrade the OS and get.

I confirm that after upgrading the Linux kernel to `6.1.0`, I obtained improved performance: ```bash sbert took: 0.27169055938720704, warmup: 0.2765231132507324 optimized sbert took: 0.11300868988037109, warmup: 0.11607575416564941 ``` @zhuhaozhe Is the...

@zhuhaozhe Yes, the full outputs are below (the first block is without ipex). By the way, could you please tell me if the `WARNING - calling in ipex numpy which...

@zhuhaozhe Thanks a lot for your reply! Could you please tell me what details do you need to have an accurate assessment? As you can see from the code, the...