failed attempt to write micrograd_from_scratch from karpathy's video in cpp :')
there's some issue with backward function, as of now, you have to declare a separate Value variable for each operation, can't do like
Value o = ((2*n).exp() - 1)/((2*n).exp() + 1)
// which is formula for tanh non linearity
it's another point that even doing this in elementary operations, doesn't provide the correct results.
also operator overloading for dtype op Value isn't providing with correct grads
mlp, layer and neuron classes are working, but due to the issue in grads, i couldn't use them to train a nueral network.
leaving this as it is for now, will check again later someday, hopefully :)