ttnn.div_bw

ttnn.div_bw(grad_tensor: ComplexTensor or ttnn.Tensor, input_tensor_a: ComplexTensor or ttnn.Tensor, input_tensor_b: ComplexTensor or ttnn.Tensor or Number, *, are_required_outputs: List[bool] = [True, True], rounding_mode: string = None, memory_config: ttnn.MemoryConfig = None, input_grad: ttnn.Tensor = None, other_grad: ttnn.Tensor = None) → None

Performs backward operations for divide on input_tensor, alpha or input_tensor_a, input_tensor_b, rounding_mode, with given grad_tensor. Supports broadcasting.

Parameters:

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

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

  • input_tensor_b (ComplexTensor or ttnn.Tensor or Number) – the input tensor.

Keyword Arguments:

  • are_required_outputs (List[bool], optional) – List of required outputs. Defaults to [True, True].

  • rounding_mode (string, optional) – Rounding mode. Defaults to None.

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

  • input_grad (ttnn.Tensor, optional) – Preallocated output tensor for gradient of input_tensor_a. Defaults to None.

  • other_grad (ttnn.Tensor, optional) – Preallocated output tensor for gradient of input_tensor_b. Defaults to None.

Note

Supported dtypes and layouts:

Dtypes

Layouts

BFLOAT16, BFLOAT8_B

TILE, ROW_MAJOR

bfloat8_b/bfloat4_b is only supported on TILE_LAYOUT

Example

# Create gradient and input tensors for division backward
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, 2], [3, 4]], 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
)
output = ttnn.div_bw(grad_tensor, tensor1, tensor2, rounding_mode=None)
logger.info(f"Division backward result: {output}")

# Create gradient and input tensors for division backward with tensor-scalar
grad_tensor = ttnn.from_torch(
    torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16), layout=ttnn.TILE_LAYOUT, device=device
)
tensor = ttnn.from_torch(
    torch.tensor([[1, 2], [3, 4]], dtype=torch.bfloat16, requires_grad=True), layout=ttnn.TILE_LAYOUT, device=device
)
scalar = 2
output = ttnn.div_bw(grad_tensor, tensor, scalar, rounding_mode=None)
logger.info(f"Division backward with tensor-scalar result: {output}")