Large language models (LLMs) have gained extended context windows through scaling positional encodings and lightweight continual pre-training. However, this often leads to degraded performance on short-text tasks, while the reasons for this degradation remain insufficiently explored. In this work, we identify two primary factors contributing to this issue: distribution drift in hidden states and attention scores, and catastrophic forgetting during continual pre-training. To address these challenges, we propose Long Context Pre-training with Restoration Distillation (LongReD), a novel approach designed to mitigate short-text performance degradation through minimizing the distribution discrepancy between the extended and original models. Besides training on long texts, LongReD distills the hidden state of selected layers from the original model on short texts. Additionally, LongReD also introduces a short-to-long distillation, aligning the output distribution on short texts with that on long texts by leveraging skipped positional indices. Experiments on common benchmarks demonstrate that LongReD effectively preserves the model’s short-text performance while maintaining or even enhancing its long-context abilities.

As Large Language Models (LLMs) continue to advance, their computational overhead has increased significantly. In this study, we identify notable redundancy across the layers of LLMs, where some layers contribute minimally to the overall network functionality. To quantify this, we introduce a metric called Block Influence (BI), which measures the importance of each layer based on the similarity between its input and output. Based on the observation of layer redundancy, we propose straightforward pruning methods for different tasks: ShortGPT for multiple-choice tasks and ShortGPT-gen for generative tasks. They prune redundant layers based on their BI scores. Our methods demonstrate superior performance over previous pruning methods. The ability to achieve better results through simple layer pruning, as opposed to more complex pruning techniques, suggests a high degree of redundancy across layers. We hope this work will contribute to future research for improving LLM efficiency.