Hi there. Is there any way to configure SameDiff to use some specific DataType by default? As far as I understand almost all operations (except the ones working with indices) use FLOAT as the default one. If I want to use BFLOAT16 or FLOAT16 - do I have to use casting everywhere, or could I simply let the SameDiff know my type preferences?
And another related question - should using any of the mentioned above data types decrease the performance? I tried already casting to FLOAT16 but I notices that in this case the performance got some 10-15% slowdown.