incerto.llm.TokenPerplexity# class incerto.llm.TokenPerplexity[source]# Bases: object Perplexity at each token position. Perplexity = exp(entropy). It measures how “surprised” the model is by its own prediction. Higher values indicate more uncertainty. __init__()# Methods __init__() compute(logits[, dim]) Compute perplexity per token. static compute(logits, dim=-1)[source]# Compute perplexity per token. Parameters: logits (Tensor) – Token logits of shape (…, vocab_size) dim (int) – Dimension to compute over (default: -1) Return type: Tensor Returns: Perplexity values of shape (…,)
