ttnn.softshrink_bw

ttnn.softshrink_bw(grad_tensor: ttnn.Tensor, input_tensor_a: ttnn.Tensor, *, lambd: float = 0.5, memory_config: ttnn.MemoryConfig = None) → List of ttnn.Tensor

Performs backward operations for softshrink on input_tensor, lambd, with given grad_tensor.

Parameters:

  • grad_tensor (ttnn.Tensor) – the input gradient tensor.

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

Keyword Arguments:

  • lambd (float, optional) – Lambda value for the softshrink formula . Defaults to 0.5.

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

Returns:

List of ttnn.Tensor – the output tensor.

Note

Supported dtypes, layouts, and ranks:

Dtypes

Layouts

Ranks

BFLOAT16, BFLOAT8_B

TILE

2, 3, 4

Example

# Create sample tensors for backward soft shrink operation
grad_tensor = ttnn.from_torch(
    torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16), layout=ttnn.TILE_LAYOUT, device=device
)
input_tensor = ttnn.from_torch(
    torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16, requires_grad=True), layout=ttnn.TILE_LAYOUT, device=device
)

# Call the softshrink_bw function with lambda parameter
output = ttnn.softshrink_bw(grad_tensor, input_tensor, lambd=0.5)
logger.info(f"Soft Shrink Backward: {output}")