index_fill

→Tensor

index_fill(input: Tensor, dim: int, index: Tensor, value: float)

Return a copy of input with positions index along dim set to value.

Autograd flows through the unmasked positions; filled positions receive zero gradient (they're overwritten by a constant).

inputTensor

Source tensor; not mutated.

dimint

Axis along which index addresses slices.

indexTensor

1-D integer tensor of positions along dim.

valuefloat

Scalar to write into every indexed position.

Tensor

Same shape and dtype as input; indexed slices replaced by value, others unchanged.