Pre-Training Curriculum for Multi-Token Prediction in Language Models
This addresses a bottleneck in training smaller language models with MTP, offering an incremental improvement to enhance their efficiency and output quality.
The paper tackles the problem that smaller language models struggle with multi-token prediction (MTP) by proposing a curriculum learning strategy with forward and reverse variants, resulting in improved downstream performance and generative quality, with the forward curriculum retaining self-speculative decoding benefits.
Multi-token prediction (MTP) is a recently proposed pre-training objective for language models. Rather than predicting only the next token (NTP), MTP predicts the next $k$ tokens at each prediction step, using multiple prediction heads. MTP has shown promise in improving downstream performance, inference speed, and training efficiency, particularly for large models. However, prior work has shown that smaller language models (SLMs) struggle with the MTP objective. To address this, we propose a curriculum learning strategy for MTP training, exploring two variants: a forward curriculum, which gradually increases the complexity of the pre-training objective from NTP to MTP, and a reverse curriculum, which does the opposite. Our experiments show that the forward curriculum enables SLMs to better leverage the MTP objective during pre-training, improving downstream NTP performance and generative output quality, while retaining the benefits of self-speculative decoding. The reverse curriculum achieves stronger NTP performance and output quality, but fails to provide any self-speculative decoding benefits.