ttnn.glu

ttnn.glu(input_tensor: ttnn.Tensor, dim: int = -1, *, memory_config: ttnn.MemoryConfig | None = None) → ttnn.Tensor

Applies glu to input_tensor element-wise.

Split the tensor into two parts, apply the GLU function on the second tensor, and then perform multiplication with the first tensor.

\[\mathrm{output\_tensor}_i = \verb|glu|(\mathrm{input\_tensor}_i)\]
Parameters:

  • input_tensor (ttnn.Tensor) – the input tensor.

  • dim (int) – Dimension to split input tensor. Supported only for last dimension (dim = -1 or 3). Defaults to -1.

Keyword Arguments:

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

Returns:

ttnn.Tensor – the output tensor.

Note

Supported dtypes, layouts, and ranks:

Dtypes

Layouts

Ranks

BFLOAT16, BFLOAT8_B

TILE

4

System memory is not supported.

Last dimension of input tensor should be divisible by 64.

Example

>>> tensor = ttnn.from_torch(torch.rand([1, 1, 32, 64], dtype=torch.bfloat16), layout=ttnn.TILE_LAYOUT, device=device)
>>> dim = 3
>>> output = ttnn.glu(tensor, dim)