I want to infer in 16bit in pytorch mkl mode after training in 32bit in pytorch gpu mode.
Is it possible?
And if inference is made in this way, will 16-bit operation speed be twice as fast as 32-bit?
I want to infer in 16bit in pytorch mkl mode after training in 32bit in pytorch gpu mode.
Is it possible?
And if inference is made in this way, will 16-bit operation speed be twice as fast as 32-bit?