CLEX: Continuous Length Extrapolation for Large Language Models

TL;DR

CLEX introduces a novel continuous length extrapolation method for LLMs that models context length scaling as differential equations, enabling effective extension of context windows beyond training lengths with minimal performance loss.

Contribution

The paper proposes CLEX, a continuous length extrapolation technique that generalizes position embedding scaling via differential equations, allowing LLMs to handle much longer contexts beyond training lengths.

Findings

CLEX extends context windows to over 4x or nearly 8x training length.

CLEX maintains performance without deterioration when extrapolating to longer contexts.

CLEX achieves competitive results on LongBench with models trained on 4k length.

Abstract

Transformer-based Large Language Models (LLMs) are pioneering advances in many natural language processing tasks, however, their exceptional capabilities are restricted within the preset context window of Transformer. Position Embedding (PE) scaling methods, while effective in extending the context window to a specific length, demonstrate either notable limitations in their extrapolation abilities or sacrificing partial performance within the context window. Length extrapolation methods, although theoretically capable of extending the context window beyond the training sequence length, often underperform in practical long-context applications. To address these challenges, we propose Continuous Length EXtrapolation (CLEX) for LLMs. We generalise the PE scaling approaches to model the continuous dynamics by ordinary differential equations over the length scaling factor, thereby overcoming the constraints of current PE scaling methods designed for specific lengths. Moreover, by extending the dynamics to desired context lengths beyond the training sequence length, CLEX facilitates the length extrapolation with impressive performance in practical tasks. We demonstrate that CLEX can be seamlessly incorporated into LLMs equipped with Rotary Position Embedding, such as LLaMA and GPT-NeoX, with negligible impact on training and inference latency. Experimental results reveal that CLEX can effectively extend the context window to over 4x or almost 8x training length, with no deterioration in performance. Furthermore, when evaluated on the practical LongBench benchmark, our model trained on a 4k length exhibits competitive performance against state-of-the-art open-source models trained on context lengths up to 32k. Our code is available at https://github.com/DAMO-NLP-SG/CLEX.