Open
Description
Creating a tensor from a float is causing a problem with the decimals:
Nx.tensor(0.43)
#Nx.Tensor<
f32
0.4300000071525574
>
The same goes if type is set to f16:
Nx.tensor(0.47, type: {:f, 16})
#Nx.Tensor<
f16
0.469970703125
>
But for f64 it seems to work:
Nx.tensor(0.47, type: {:f, 64})
#Nx.Tensor<
f64
0.47
>