Adapt test CI to run tests with multiple torch versions, including 1.13
Add _grad_input_padding function that was removed in 1.13 as utility and import if the installed torch version doesn't include it anymore
parameters_to_vector stopped working for tensor lists with non-contiguous entries. Fix by calling contiguous() before vectorizing.
torch raises a new exception (TypeError) when trying to compute the Hessian of a linear function. Fix by adding the new exception to the catched exceptions.
Fixes #272.
_grad_input_padding
function that was removed in 1.13 as utility and import if the installed torch version doesn't include it anymoreparameters_to_vector
stopped working for tensor lists with non-contiguous entries. Fix by callingcontiguous()
before vectorizing.TypeError
) when trying to compute the Hessian of a linear function. Fix by adding the new exception to the catched exceptions.