Tensor.scaled_uniform()

Generates a tensor of random values from a uniform distribution and scales the tensor by a factor proportional to the square root of the product of the tensor's shape dimensions.

Usage

from tinygrad.tensor import Tensor
 
tensor = Tensor.scaled_uniform(3, 3)
 
print(tensor.numpy())

Return value

[[ 0.05666578  0.32397062  0.0690138 ]
 [ 0.05537931  0.18648589 -0.24747562]
 [-0.11055386 -0.1891321  -0.16025254]]

Scaled Dot Product Attention Shape

;