Difference between dtype and precision arguments?

[amishra791]

Hi, I was looking at the documentation for the Dense layer and I wanted to clarify my understanding of the two arguments.

To my understanding, precision determines the number of bfloat16 passes for float32 computations. So if dtype was anything other than float32, precision would have no effect, correct? Also, in case dtype wasn't float32 (e.g. bfloat16 or even lower), will all computations done by lax will be in the same dtype?