I'm done trying to predict output types. To simplify the ACE codes especially with an eye to creating more flexible model building and utilizing chains I've removed all instances of valtype and gradtype.
The cost is that we have some additional allocations but so far it looks they don't cost all that much. We will continue to benchmark and see where we can improve.
cortner
commented
2 years ago
I'm done trying to predict output types. To simplify the ACE codes especially with an eye to creating more flexible model building and utilizing chains I've removed all instances of valtype and gradtype.
The cost is that we have some additional allocations but so far it looks they don't cost all that much. We will continue to benchmark and see where we can improve.