Fix 341 (#346)
* split greedy and sampling generative + remove small old helm mechanism
* add do_sample to generative tas criteria
* Quick fix vllm (#361)
* fix max len management in vllm
* fixed the maj@n qem being run on the same samples. needed to manage the sort and split
* add temperature to vllm config
---------
Co-authored-by: Nathan Habib <nathan.habib@huggingface.co>
Co-authored-by: Nathan Habib <30601243+NathanHB@users.noreply.github.com>
