ttnn.randn

ttnn.randn(shape: list[int], *, device: ttnn.Device | ttnn.MeshDevice = None, dtype: ttnn.DataType = ttnn.bfloat16, layout: ttnn.Layout = ttnn.TILE_LAYOUT, memory_config: ttnn.MemoryConfig = ttnn.DRAM_MEMORY_CONFIG, compute_kernel_config: ttnn.DeviceComputeKernelConfig = None, seed: int = None) → ttnn.Tensor

Generates a tensor with the given shape, filled with random values from a standard normal distribution. Internally, this operation uses the Box-Muller transform to generate normally distributed random values.

Parameters:

shape (list[int]) – a list of integers defining the shape of the output tensor.

Keyword Arguments:

  • device (ttnn.Device | ttnn.MeshDevice, optional) – The device on which the tensor will be allocated. Defaults to None.

  • dtype (ttnn.DataType, optional) – The data type of the tensor. Defaults to ttnn.bfloat16.

  • layout (ttnn.Layout, optional) – The layout of the tensor. Defaults to ttnn.TILE_LAYOUT.

  • memory_config (ttnn.MemoryConfig, optional) – Memory configuration for the operation. Defaults to ttnn.DRAM_MEMORY_CONFIG.

  • compute_kernel_config (ttnn.DeviceComputeKernelConfig, optional) – Configuration for the compute kernel. Defaults to None.

  • seed (int, optional) – An optional seed to initialize the random number generator for reproducible results. Defaults to None.

Returns:

ttnn.Tensor – the output tensor.

Note

The output tensor supports the following data types and layouts:

Output Tensor

dtype

layout

FLOAT32

ROW_MAJOR, TILE

BFLOAT16

ROW_MAJOR, TILE

Note

Memory layout support: interleaved DRAM and L1.

Example

# Create a TT-NN tensor with random values standard normal distributed
tensor = ttnn.randn(shape=[2, 3], dtype=ttnn.bfloat16, layout=ttnn.ROW_MAJOR_LAYOUT, device=device)
logger.info("TT-NN randn tensor:", tensor)