165. Perplexity

Given a list of per-token negative log-likelihoods (natural log), compute the perplexity of a language model's predictions.

Signature: def perplexity(neg_log_likelihoods: list) -> float

Perplexity is the exponential of the average negative log-likelihood (see the math reference). The output is a single Python float. Reduce over all elements of the input — for 2D inputs of shape (B, T) the mean is over both batch and time axes, not just one.

Note: NLLs are assumed to use the natural logarithm (base e).