torch.tx.pipePatterns

function pipePatterns(...patterns: string[]): <T extends Record<string, TensorOrNumber>>(tensors: T) => AnyTensor

Pipe multiple patterns together.

Each pattern's result becomes the _ input for the next pattern. This creates a pipeline of transformations, similar to function composition.

Parameters

patternsstring[]: Patterns to pipe together (first is innermost)

Returns

<T extends Record<string, TensorOrNumber>>(tensors: T) => AnyTensor– A function that evaluates the piped expression

Examples

// Build a two-layer MLP: sigmoid(relu(x @ w1 + b1) @ w2 + b2)
const mlp = $.pipe(
  'x @ w1 + b1',      // First linear layer
  'relu(_)',          // ReLU activation
  '_ @ w2 + b2',      // Second linear layer
  'sigmoid(_)'        // Output activation
);

const output = mlp({ x, w1, b1, w2, b2 });