Is your feature request related to a problem? Please describe.
When reading the nodes in the JSON if there is consecutive nodes: MatMul -> BiasAdd -> ActivationFunction then the framework should instead of creating three different object create a single FusedLinear where the full operation can be optimized more easily.
Describe the solution you'd like
A implementation of the Layer class which implements the logic for computing the full operation from end-to-end.
Describe alternatives you've considered
An alternative would be creating separate object which handle one operation at a time, this would work but hinders the framework from utilizing more optimized solutions like gemm which does the matrix multiplication and bias at the same time with a significant speedup.
Additional context
Not currently.
Is your feature request related to a problem? Please describe.
When reading the nodes in the JSON if there is consecutive nodes: MatMul -> BiasAdd -> ActivationFunction then the framework should instead of creating three different object create a single FusedLinear where the full operation can be optimized more easily.
Describe the solution you'd like
A implementation of the Layer class which implements the logic for computing the full operation from end-to-end.
Describe alternatives you've considered
An alternative would be creating separate object which handle one operation at a time, this would work but hinders the framework from utilizing more optimized solutions like gemm which does the matrix multiplication and bias at the same time with a significant speedup.
Additional context
Not currently.