LongAlign: A Recipe for Long Context Alignment of Large Language Models

Extending large language models to effectively handle long contexts requires instruction finetuning on input sequences of similar length. To address this, we present LongAlign-a recipe of the instruction data, training, and evaluation for long context alignment. First, we construct a long instruction-following dataset using Self-Instruct. To ensure the data diversity, it covers a broad range of tasks from various long context sources. Second, we investigate different strategies to speed up supervised fine-tuning on datasets with uneven length distribution, namely packing and sorted batching. Additionally, we develop a loss weighting method to balance the contribution to the loss across different sequences during packing training. Third, we introduce the LongBench-Chat benchmark for evaluating instruction-following capabilities on queries of 10k-100k in length. Experiments show that LongAlign outperforms existing recipes for LLMs in long context tasks by up to 30%, while also maintaining their proficiency in handling short, generic tasks. The code, data, and long-aligned models are open-sourced at https://github.com/ THUDM/LongAlign.