[FIX] fix run_generation.py to work with batch_size > 1

[mataney]

I expended the top_k_top_p_filtering function, and by that therun_generation.py script to work with num_samples > 1.

This can be expended by scattering the sorted tensors.
First pull request in this repository, so let me know if I need to do anything else :)

Cheers, Matan.

[mataney]

@thomwolf
I created this PR to deal with the top p generations.
Should I have opened an issue first to check if it is needed?
Should I deal with the conflicts?

Cheers.

[julien-c]

Hi @mataney, thanks.

This was rebased, fixed by f96ce1c, and merged to master by 2a5663c.

Check that everything looks good on your side if you can.