Unique function for float tensors

hello.
I have a float tensor and i am welling to run torch.unique on it , however it does not work but if i convert it to int or long it does work, i know that the different float values in the tensor are finite because im quantizing to some bit, any idea how can i run torch.unique on float tensor?
Thanks alot!

a=torch.tensor([1., 1.+1e-7, 1.+1e-6, 1.+1e-5, 1.5])
explvl=torch.tensor([20]) #threshold 1/(2**20) ~= 1e-6
a.ldexp(explvl).round().unique().ldexp(-explvl).tolist()
: [1.0, 1.000_0009536743164, 1.000_009536743164, 1.5]