LongRecipe: Recipe for Efficient Long Context Generalization in Large Language Models

TL;DR

LongRecipe is an efficient training method that significantly extends the context window of large language models, enabling better long-range dependency understanding with reduced computational resources.

Contribution

We propose LongRecipe, a novel training strategy that extends LLMs' context window efficiently without extensive retraining, improving long-sequence processing capabilities.

Findings

Extends context window from 8k to 128k tokens.

Reduces training resources by over 85%.

Maintains performance on general tasks.

Abstract

Large language models (LLMs) face significant challenges in handling long-context tasks because of their limited effective context window size during pretraining, which restricts their ability to generalize over extended sequences. Meanwhile, extending the context window in LLMs through post-pretraining is highly resource-intensive. To address this, we introduce LongRecipe, an efficient training strategy for extending the context window of LLMs, including impactful token analysis, position index transformation, and training optimization strategies. It simulates long-sequence inputs while maintaining training efficiency and significantly improves the model's understanding of long-range dependencies. Experiments on three types of LLMs show that LongRecipe can utilize long sequences while requiring only 30% of the target context window size, and reduces computational training resource over 85% compared to full sequence training. Furthermore, LongRecipe also preserves the original LLM's capabilities in general tasks. Ultimately, we can extend the effective context window of open-source LLMs from 8k to 128k, achieving performance close to GPT-4 with just one day of dedicated training using a single GPU with 80G memory. Our code is released at https://github.com/zhiyuanhubj/LongRecipe.