ttnn.typecast

ttnn.typecast() → ttnn.Tensor

Applies typecast to input_tensor.

:param * input_tensor: input tensors must be on device, in ROW MAJOR or TILE layout :type * input_tensor: ttnn.Tensor :param * dtype: data type must be one of the following types BFLOAT16,BFLOAT8_B,BFLOAT4_B,UINT32,INT32 and UINT16. :type * dtype: Optional[ttnn.DataType] :param *:

:keyword * memory_config: Memory configuration for the operation. :kwtype * memory_config: Optional[ttnn.MemoryConfig] :keyword * output_tensor: Preallocated tensor to store the output. :kwtype * output_tensor: Optional[ttnn.Tensor]

Returns:

ttnn.Tensor – The tensor with the updated data type. Output tensor will be on device, in same layout, and have the given data type.

Example:

>>> tensor = ttnn.typecast(torch.randn((10, 3, 32, 32), dtype=ttnn.bfloat16), ttnn.uint16)