Mask-Enhanced Autoregressive Prediction: Pay Less Attention to Learn More
This addresses a critical bottleneck in large language models for applications requiring precise information retrieval, though it appears incremental as it combines existing techniques.
The paper tackles the problem of large language models struggling with accurate key information retrieval by proposing Mask-Enhanced Autoregressive Prediction (MEAP), which integrates masked language modeling into next-token prediction to enhance retrieval capabilities without extra computational cost. Results show MEAP substantially outperforms standard training on key information retrieval and long-context reasoning, with an 11.77 percentage point advantage in lost-in-the-middle scenarios during fine-tuning.
Large Language Models (LLMs) are discovered to suffer from accurately retrieving key information. To address this, we propose Mask-Enhanced Autoregressive Prediction (MEAP), a simple yet effective training paradigm that seamlessly integrates Masked Language Modeling (MLM) into Next-Token Prediction (NTP) to enhance the latter's in-context retrieval capabilities. Specifically, MEAP first randomly masks a small fraction of input tokens and then directly performs the standard next-token prediction autoregressive using a decoder-only Transformer. MEAP eliminates the need for bidirectional attention or encoder-decoder architectures for MLM, incurring no additional computational overhead during pre-training or inference. Intensive experiments demonstrate that MEAP substantially outperforms NTP on key information retrieval and long-context reasoning tasks, while performing on par or better on commonsense reasoning tasks. The benefits of MEAP also extend to supervised fine-tuning, where it shows remarkable advantages in lost-in-the-middle scenarios, outperforming NTP by 11.77 percentage points. Our analysis indicates that MEAP's effectiveness arises from its ability to promote more distinguishable attention scores by concentrating on a reduced set of non-masked tokens. This mechanism improves the model's focus on task-relevant signals while mitigating the influence of peripheral context. These findings position MEAP as a promising training paradigm for large language models.