ttnn.mul_bw

ttnn.mul_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], memory_config: ttnn.MemoryConfig = None, input_grad: ttnn.Tensor = None, other_grad: ttnn.Tensor = None) → List of ttnn.Tensor

Performs backward operations for multiply on input_tensor_a, input_tensor_b, with given grad_tensor.

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].

  • 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.

Returns:

List of ttnn.Tensorgradients shaped as their respective operands (PyTorch parity). Preallocated input_grad/other_grad must match input_tensor_a/input_tensor_b shape, not grad_tensor — differs from the pre-#56601 grad-shape contract on broadcast.

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 multiplication 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
)

# Compute gradients for multiplication operation
output = ttnn.mul_bw(grad_tensor, tensor1, tensor2)
logger.info(f"Multiplication backward result: {output}")

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
)
scalar = 4
output = ttnn.mul_bw(grad_tensor, tensor1, scalar)
logger.info(f"Multiplication backward with scalar result: {output}")