ttnn.to_dtype

ttnn.to_dtype(tensor: ttnn.Tensor, dtype: ttnn.DataType) → None

Converts a host tensor to the desired dtype.

Parameters:

  • tensor (ttnn.Tensor) – the tensor to be converted to the desired dtype.

  • dtype (ttnn.DataType) – the desired data type.

Note

This operations supports tensors according to the following data types and layout:

tensor

dtype - layout

BFLOAT16, BFLOAT8_B, BFLOAT4_B, FLOAT32, UINT32, INT32, UINT16, UINT8 - TILE

BFLOAT16, FLOAT32, UINT32, INT32, UINT16, UINT8 - ROW_MAJOR
Memory Support:

  • Interleaved: DRAM and L1

  • Height, Width, Block, and ND Sharded: DRAM and L1

Limitations:

  • tensor must be on the host.

Example

# Create a TT-NN tensor on the host and convert it to a different data type
tensor = ttnn.rand((10, 64, 32), device=device, dtype=ttnn.bfloat16)
tensor = ttnn.from_device(tensor)  # to_dtype requires a host tensor
tensor = ttnn.to_dtype(tensor, dtype=ttnn.uint8)
assert tensor.dtype == ttnn.uint8
assert tensor.shape == (10, 64, 32)

logger.info("TT-NN tensor shape after converting to uint8", tensor.shape)