Skip to main content

Fork

Fork (Duplicate)

Splits a single input into two identical outputs. Essential building block for residual connections, parallel processing paths, and skip connections.

Shape Contract:

  • Input: [*shape] arbitrary shape
  • Output main: [*shape] first copy of input
  • Output skip: [*shape] second copy of input

Notes:

  • No learnable parameters (pure structural operation)
  • Zero-copy in most implementations (shares memory)
  • Named outputs (main, skip) for clarity in residual patterns
  • Both outputs are identical references to the input
  • Fundamental for any architecture with parallel paths

Signature​

neuron Fork()

Ports​

Inputs:

  • default: [*shape]

Outputs:

  • main: [*shape]
  • skip: [*shape]

Implementation​

Source { source: "core", path: "structural/Fork" }