full_like

→Tensor

full_like(input: Tensor, fill_value: _complex, dtype: DTypeLike = None, device: DeviceLike = None, requires_grad: _bool = False)

source edit

Implementing kernel

C++full_like_opfree fn

Return a constant-filled tensor with the same shape, dtype, and device as input.

Every element of the output equals fill_value:

F_{i_1, \ldots, i_n} = c \quad \forall\; (i_1, \ldots, i_n) \in \prod_k [0, s_k)

where $c = \texttt{fill\_value}$ and $(s_1, \ldots, s_n) = \texttt{t.shape}$ . This is the shape-aware generalisation of full.

Parameters

inputTensor

Reference tensor. Shape, dtype, and device are inherited unless overridden by the keyword arguments.

fill_valuebool, int, float or complex

Scalar constant to broadcast across all elements, stored exactly in the result's dtype as full stores it.

dtypelucid.dtype

Override the data type. The value is filled in this dtype directly, not filled in t.dtype and cast — so full_like(int_tensor, 2.5, dtype=lucid.float32) holds 2.5.

devicestr or lucid.device

Override the device.

requires_gradbool

If True, the constant tensor is a leaf recorded by autograd. Default: False.

Returns

Tensor

Constant tensor shaped like input.

Notes

A canonical application is initialising the attention bias mask before selectively unmasking positions. Starting from $-\infty$ on a tensor shaped like the attention weight matrix ensures that masked positions become zero after softmax:

\text{logit}_{ij} \leftarrow \begin{cases} \text{logit}_{ij} & \text{unmasked} \\ -\infty & \text{masked} \end{cases} \implies \text{softmax}(\text{logit})_i \;=\; 0 \text{ for masked } j

Examples

>>> import lucid
>>> scores = lucid.randn(4, 8)
>>> mask = lucid.full_like(scores, float("-inf"))
>>> mask.shape
(4, 8)
Constant padding value:
>>> x = lucid.randn(2, 3)
>>> padded = lucid.full_like(x, -1.0)

>>> import lucid >>> scores = lucid.randn(4, 8) >>> mask = lucid.full_like(scores, float("-inf")) >>> mask.shape (4, 8) Constant padding value: >>> x = lucid.randn(2, 3) >>> padded = lucid.full_like(x, -1.0)