Comment on Scalable MatMul-Free Language ModelingComments−PaulHoule2yThis is why the NPU built into your processor could quickly become a liability instead of a benefit.−imtringued2yMost NPUs advertise TOPS when most models still run on floating point. The approach they outlined will also run faster on NPUs.−sva_2yIt's ok, the AMD one still doesn't even work on Linux without a custom kernel
Comments
This is why the NPU built into your processor could quickly become a liability instead of a benefit.
Most NPUs advertise TOPS when most models still run on floating point. The approach they outlined will also run faster on NPUs.
It's ok, the AMD one still doesn't even work on Linux without a custom kernel