ttnn.softplus

ttnn.softplus(input_tensor: ttnn.Tensor, *, beta: float | None = 1, threshold: float | None = 20, memory_config: ttnn.MemoryConfig | None = None, output_tensor: ttnn.Tensor | None = None, queue_id: int | None = 0) → ttnn.Tensor

Applies softplus to input_tensor element-wise.

\[\mathrm{output\_tensor}_i = softplus(\mathrm{input\_tensor}_i)\]
Parameters:

input_tensor (ttnn.Tensor) – the input tensor.

Keyword Arguments:

  • beta (float, optional) – Scales the input before applying the Softplus function. By modifying beta, you can adjust the steepness of the function. A higher beta value makes the function steeper, approaching a hard threshold like the ReLU function for large values of beta. Defaults to 1.

  • threshold (float, optional) – Used to switch to a linear function for large values to improve numerical stability. This avoids issues with floating-point representation for very large values. Defaults to 20.

  • memory_config (ttnn.MemoryConfig, optional) – Memory configuration for the operation. Defaults to None.

  • output_tensor (ttnn.Tensor, optional) – preallocated output tensor. Defaults to None.

  • queue_id (int, optional) – command queue id. Defaults to 0.

Returns:

ttnn.Tensor – the output tensor.

Note

Supported dtypes, layouts, and ranks:

Dtypes

Layouts

Ranks

BFLOAT16

TILE

2, 3, 4

Example

>>> tensor = ttnn.from_torch(torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16), dtype=ttnn.bfloat16, layout=ttnn.TILE_LAYOUT, device=device)
>>> output = ttnn.softplus(tensor, beta = 1.0, threshold = 20.0)