ttnn.format_input_tensor

ttnn.format_input_tensor(input: ttnn._ttnn.tensor.Tensor, device: ttnn._ttnn.device.IDevice, padded_shape: tt::tt_metal::SimpleShape, pad_value: float, target_layout: ttnn._ttnn.tensor.Layout, target_mem_config: Optional[ttnn._ttnn.tensor.MemoryConfig] = None) → ttnn._ttnn.tensor.Tensor

Formats tensor to target layout and pads to padded shape.

Parameters:

  • input (ttnn.Tensor) – Input tensor to format.

  • device (ttnn.device.Device) – Device where the tensor will be moved.

  • padded_shape (ttnn.Shape) – Desired shape of the tensor.

  • pad_value (float) – Value to pad with.

  • target_layout (ttnn.Layout) – Desired tensor layout.

  • target_mem_config (ttnn.MemoryConfig, optional) – Desired memory config. Defaults to None.

Returns:

ttnn.Tensor – Formatted tensor.

Note

This functionality is planned for deprecation in the future.

Example

>>> input_tensor = ttnn.ones([1, 2, 2, 2], dtype=ttnn.bfloat16, layout=ttnn.ROW_MAJOR_LAYOUT)
>>> padded_tensor = ttnn.format_input_tensor(input_tensor, device=device, padded_shape=[1, 2, 4, 4], pad_value=0.0, target_layout=ttnn.TILE_LAYOUT, output_mem_config)