ppl : fix n_seq_max for perplexity

[slaren]

The value of n_parallel (= n_seq_max) was being overwritten for hellaswag, which can result in the wrong rope factors being used with Phi3.

[ngxson]

LGTM.

Nitpicks, but I'd prefer to reverse the bool ppl condition to be more readable:

const bool not_ppl = params.hellaswag || params.winogrande || params.multiple_choice || params.kl_divergence;

if (not_ppl) {
    ...
} else {
    ...
}