Eggroll and Evolution Strategy Implementation
This implementation literally use (x @ A) @ B.T to avoid large matrix calculation
x = nn.functional.linear(input, self.a_weight) # in -> rank
x = nn.functional.linear(x, self.b_weight, self.bias) # rank -> out