A fix for the memory leak I describe in issue 284. Specifically, when using KeOps with ranges but no batch dimension, slices_x_d and ranges_y_d are not freed.
After checking, the error was caused by the fact that my tensor were on CPU, but this is still a weird behavior to have a memory leak on GPU when the computations should be on CPU.
A fix for the memory leak I describe in issue 284. Specifically, when using KeOps with ranges but no batch dimension,
slices_x_d
andranges_y_d
are not freed.