ttnn.repeat_bw

ttnn.repeat_bw(grad_tensor: ttnn.Tensor, input_tensor: ttnn.Tensor, shape: List[int], *, memory_config: ttnn.MemoryConfig = None) → List of ttnn.Tensor

Performs backward operations for repeat on input_tensor, with given grad_tensor using given shape.

Parameters:

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

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

• shape (List[int]) – Shape of tensor.

Keyword Arguments:

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	4

Example

# Create sample tensors for backward repeat operation
# Input tensor that will be repeated
input_tensor = ttnn.from_torch(
    torch.rand([1, 1, 32, 32], dtype=torch.bfloat16, requires_grad=True), layout=ttnn.TILE_LAYOUT, device=device
)
# Grad tensor matching the shape after repeat operation (2x repeat in dim 0)
grad_tensor = ttnn.from_torch(
    torch.rand([2, 1, 32, 32], dtype=torch.bfloat16), layout=ttnn.TILE_LAYOUT, device=device
)
# Define the shape for repeat operation
shape = [2, 1, 1, 1]

# Call the repeat_bw function
output = ttnn.repeat_bw(grad_tensor, input_tensor, shape)
logger.info(f"Repeat Backward: {output}")