ttnn.rdiv_bw

ttnn.rdiv_bw(grad_tensor: ttnn.Tensor, input_tensor_a: ttnn.Tensor, scalar: float, *, round_mode: string = None, memory_config: ttnn.MemoryConfig = None) → List of ttnn.Tensor

Performs backward operations for Unary rdiv on input_tensor, scalar with given grad_tensor using given round_mode. round_mode can be ‘None’, ‘trunc’, or ‘floor’.

Parameters:

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

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

  • scalar (float) – divisor.

Keyword Arguments:

  • round_mode (string, optional) – Mode of Rounding. 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

Performance of the PCC may degrade when using BFLOAT8_B. For more details, refer to the BFLOAT8_B limitations.

Example

# Create sample tensors for backward reverse division 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
)
# Define scalar value for reverse division
scalar = 0.5

# Call the rdiv_bw function with scalar and round mode
output = ttnn.rdiv_bw(grad_tensor, input_tensor, scalar, round_mode=None)
logger.info(f"Reverse Division Backward: {output}")