I have doubt regarding the inference in the quantization mobile devices.
Does it do all the operations in int8 weights or the converted int8 weights will convert into float32 and then inference will run on float32 weights ?
I have doubt regarding the inference in the quantization mobile devices.
Does it do all the operations in int8 weights or the converted int8 weights will convert into float32 and then inference will run on float32 weights ?