A pair of pure functions implementing a gradient transformation.
Optimizers are implemented with this interface. They do not contain any
internal state. The "optimizer state" OptState is initialized, then passed
into each update call, which returns a new state.
Gradients are transformed during the update call, and they should have the
same PyTree shape as the parameters.
A pair of pure functions implementing a gradient transformation.
Optimizers are implemented with this interface. They do not contain any internal state. The "optimizer state"
OptStateis initialized, then passed into each update call, which returns a new state.Gradients are transformed during the update call, and they should have the same PyTree shape as the parameters.