多令牌预测需要寄存器

Multi-Token Prediction Needs Registers

May 15, 2025

作者: Anastasios Gerontopoulos, Spyros Gidaris, Nikos Komodakis

cs.AI

摘要

多令牌预测作为一种提升语言模型预训练效果的有前景目标已崭露头角，但其优势尚未在诸如微调等其他场景中普遍显现。本文提出MuToR，一种简单而有效的多令牌预测方法，它将可学习的寄存器令牌交错插入输入序列，每个令牌负责预测未来的目标。与现有方法相比，MuToR具备几大关键优势：仅引入极少量的额外参数，无需架构改动——确保与现成预训练语言模型的兼容性——且与下一令牌预训练目标保持一致，使其特别适用于监督微调。此外，它天然支持可扩展的预测范围。我们通过一系列用例，包括监督微调、参数高效微调（PEFT）及预训练，在语言和视觉领域的挑战性生成任务上，展示了MuToR的有效性与多功能性。我们的代码将发布于：https://github.com/nasosger/MuToR。

English

Multi-token prediction has emerged as a promising objective for improving language model pretraining, but its benefits have not consistently generalized to other settings such as fine-tuning. In this paper, we propose MuToR, a simple and effective approach to multi-token prediction that interleaves learnable register tokens into the input sequence, each tasked with predicting future targets. Compared to existing methods, MuToR offers several key advantages: it introduces only a negligible number of additional parameters, requires no architectural changes--ensuring compatibility with off-the-shelf pretrained language models--and remains aligned with the next-token pretraining objective, making it especially well-suited for supervised fine-tuning. Moreover, it naturally supports scalable prediction horizons. We demonstrate the effectiveness and versatility of MuToR across a range of use cases, including supervised fine-tuning, parameter-efficient fine-tuning (PEFT), and pretraining, on challenging generative tasks in both language and vision domains. Our code will be available at: https://github.com/nasosger/MuToR.