ttnn.clamp_bw

ttnn.clamp_bw(grad_tensor: ttnn.Tensor, input_tensor: ttnn.Tensor, *, min: float = None, max: float = None, memory_config: ttnn.MemoryConfig = None) → List of ttnn.Tensor

Performs backward operations for clamp on input_tensor, min, max with given grad_tensor.

Parameters:

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

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

Keyword Arguments:

  • min (float, optional) – Minimum value. Defaults to None.

  • max (float, optional) – Maximum value. Defaults to None.

  • 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

TILE

2, 3, 4

Only one of min or max value can be None.

Example

# Create sample tensors
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
)
# Define min and max values for clamping
min_val = 0.5
max_val = 2.0

# Call the clamp_bw function
output = ttnn.clamp_bw(grad_tensor, input_tensor, min_val, max_val)
logger.info(f"Clamped Output Backward: {output}")