torch.js has not been released yet.

torch.index_put

function index_put(input: Tensor, indices: (Tensor | null)[], values: Tensor, options?: IndexPutOptions): Tensor

function index_put(input: Tensor, indices: (Tensor | null)[], values: Tensor, accumulate: boolean, options?: IndexPutOptions): Tensor

Places values into a tensor at index positions (general indexing operation).

General form of index assignment supporting multiple index tensors or None. Currently supports simplified single-dimension indexing. Equivalent to: input[index] = values for single index tensors. Useful for:

Multi-dimensional indexing: setting values at complex positions
Conditional assignment: setting based on computed indices
General index_put replacement: Python-style tensor[indices] = values
Flexible placement: placing at arbitrary positions

Note: Currently only supports single dimension indexing. Use index_copy or scatter operations for more complex indexing patterns.

Single dimension: Currently limited to single 1D index tensor
Accumulate flag: Controls replacement vs addition behavior
Index restrictions: Only supports simple 1D indexing for now

Limited scope: Multi-dimensional advanced indexing not yet fully supported
Index type: Only 1D index tensors currently supported

Parameters

inputTensor: The input tensor
indices(Tensor | null)[]: Array of index tensors or None (currently only 1D single index supported)
valuesTensor: Source values to place
optionsIndexPutOptionsoptional

Returns

Tensor– A new tensor with values placed at index positions

Examples

// Simple 1D indexing
const data = torch.zeros(10);
const indices = [torch.tensor([0, 2, 4, 6, 8])];
const values = torch.ones(5);
const result = torch.index_put(data, indices, values);  // [0]=1, [2]=1, etc.

// Accumulate mode
const accumulator = torch.zeros(10);
const acc_indices = [torch.tensor([0, 0, 1, 1, 2])];  // Repeated indices
const acc_values = torch.ones(5);
const accumulated = torch.index_put(accumulator, acc_indices, acc_values, true);
// [0]=2, [1]=2, [2]=1 (accumulated at same positions)

See Also