why nnx.split_rngs as decorator used in __call__

[A373k]

class Block(nnx.Module):
  def __init__(self, input_dim, features, rngs):
    self.linear = nnx.Linear(input_dim, features, rngs=rngs)
    self.dropout = nnx.Dropout(0.5, rngs=rngs)

  def __call__(self, x: jax.Array):  # No need to require a second input!
    x = self.linear(x)
    x = self.dropout(x)
    x = jax.nn.relu(x)
    return x   # No need to return a second output!

class MLP(nnx.Module):
  def __init__(self, features, num_layers, rngs):
    @nnx.split_rngs(splits=num_layers)
    @nnx.vmap(in_axes=(0,), out_axes=0)
    def create_block(rngs: nnx.Rngs):
      return Block(features, features, rngs=rngs)

    self.blocks = create_block(rngs)
    self.num_layers = num_layers

  def __call__(self, x):
    @nnx.split_rngs(splits=self.num_layers)
    @nnx.scan(in_axes=(nnx.Carry, 0), out_axes=nnx.Carry)
    def forward(x, model):
      x = model(x)
      return x

    return forward(x, self.blocks)

model = MLP(64, num_layers=5, rngs=nnx.Rngs(0))

why @nnx.split_rngs(splits=self.num_layers) on forward function as its used in create_block function