kevinintel

Hi, I will close this issue if you don't have concerns

Hi, I will close this issue if you don't have concerns

Can you share your quntization step?

Thanks for the reminding, we will investigate into it

It's in our plan. Thanks

Without numa, the performance will drop a lot

Intel Xeon offen has 2 sockets, -m 0 aimed to bind the memory in first socket. There are overhead of communcation between 2 sockets, if you want to reduce internode,...

we will not support it wiht sycl, but we may use our kernels to support it.

@benjamin27315k, I will close this issue if you don't have concerns

@yufenglee I will close this issue if you don't have concerns