ttnn.where_bw

ttnn.where_bw(grad_tensor: ttnn.Tensor, input_tensor_a: ttnn.Tensor, input_tensor_b: ttnn.Tensor, input_tensor_c: ttnn.Tensor, *, are_required_outputs: List[bool] = [True, True], memory_config: ttnn.MemoryConfig = None, output_tensor: ttnn.Tensor = None) → List of ttnn.Tensor

Performs backward operations for where of input_tensor_a, input_tensor_b and input_tensor_c with given grad_tensor.

Parameters:

grad_tensor (ttnn.Tensor) – the input gradient tensor.
input_tensor_a (ttnn.Tensor) – the input tensor.
input_tensor_b (ttnn.Tensor) – the input tensor.
input_tensor_c (ttnn.Tensor) – the input tensor.

Keyword Arguments:

are_required_outputs (List[bool], optional) – list of required outputs. Defaults to [True, True].
memory_config (ttnn.MemoryConfig, optional) – memory configuration for the operation. Defaults to None.
output_tensor (ttnn.Tensor, optional) – preallocated output tensor. 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 three tensors and a gradient tensor for the operation
grad_tensor = ttnn.from_torch(
    torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16), layout=ttnn.TILE_LAYOUT, device=device
)
tensor1 = ttnn.from_torch(
    torch.tensor([[1, 0], [1, 0]], dtype=torch.bfloat16, requires_grad=True), layout=ttnn.TILE_LAYOUT, device=device
)
tensor2 = ttnn.from_torch(
    torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16, requires_grad=True), layout=ttnn.TILE_LAYOUT, device=device
)
tensor3 = ttnn.from_torch(
    torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16, requires_grad=True), layout=ttnn.TILE_LAYOUT, device=device
)

# Perform the where backward operation
output = ttnn.where_bw(grad_tensor, tensor1, tensor2, tensor3)
logger.info(f"Where Backward result: {output}")