ttnn.to_device

ttnn.to_device() → None

Copies the ttnn.Tensor tensor to the tt_lib.device.MeshDevice.

The tensor may be placed in DRAM or L1 memory.

Currently memory_config must be of an Interleaved tensor (not sharded)

:param * tensor: the ttnn.Tensor :param * device: the ttnn.MeshDevice :param * memory_config: the optional MemoryConfig (DRAM_MEMORY_CONFIG or L1_MEMORY_CONFIG). Defaults to DRAM_MEMORY_CONFIG.

Example

# Open the device
# device_id = 0
# device = ttnn.open_device(device_id=device_id)

# Create a TT-NN tensor and move it to the specified device
tensor_on_host = ttnn.from_torch(torch.randn((10, 64, 32)), dtype=ttnn.bfloat16)
ttnn_tensor = ttnn.to_device(tensor_on_host, device=device)

logger.info("TT-NN tensor shape after moving to device", ttnn_tensor.shape)