index_put

→Tensor

index_put(input: Tensor, indices: list[Tensor] | tuple[Tensor, ...], values: Tensor, accumulate: bool = False)

source edit

Out-of-place advanced-indexing write.

Equivalent to out = input.clone(); out[indices] = values (or out[indices] += values when accumulate=True) under reference framework semantics. indices is a sequence of integer tensors, one per leading dimension; broadcasting between them follows the standard rules.

Currently restricted to the case where len(indices) == input.ndim and every index tensor broadcasts to a common shape — partial advanced indexing (where trailing dims are implicitly sliced) is filed as a follow-up.

Parameters

inputTensor

Destination tensor.

indicessequence of Tensors

One integer index tensor per dimension of input. All broadcast to a common shape.

valuesTensor

Values to scatter, broadcastable to the common index shape.

accumulatebool= False

If True, add at each position; otherwise overwrite.