标签
This paper proposes Pruned BPE, a post-training method that prunes low-exposure tokens from a BPE vocabulary and reallocates slots to better-exposed candidates, reducing encoded length without increasing model-visible vocabulary size. Experiments on English and Chinese corpora show approximately 0.27–0.36% encoded length reduction over standard BPE.
MicroSpec 是一种无需训练的技术，它能即时构建紧凑的上下文感知词汇表，以加速大型语言模型中的推测解码，将平均词汇表大小减少40倍以上，并相比EAGLE-2实现了高达1.32倍的端到端加速。