CoPE: Clipped RoPE as A Scalable Free Lunch for Long Context LLMs

TL;DR

CoPE introduces a soft clipping method for Rotary Positional Embedding that enhances long context handling in Large Language Models, outperforming previous approaches and enabling scalable context lengths up to 256k tokens.

Contribution

This work presents CoPE, a simple yet effective soft clipping technique for RoPE that unifies OOD mitigation and semantic modeling, improving long context performance in LLMs.

Findings

Significant performance improvements with CoPE on long context tasks

Scalability of RoPE up to 256k tokens demonstrated

Theoretical analysis supports effectiveness of soft clipping

Abstract

Rotary Positional Embedding (RoPE) is a key component of context scaling in Large Language Models (LLMs). While various methods have been proposed to adapt RoPE to longer contexts, their guiding principles generally fall into two categories: (1) out-of-distribution (OOD) mitigation, which scales RoPE frequencies to accommodate unseen positions, and (2) Semantic Modeling, which posits that the attention scores computed with RoPE should always prioritize semantically similar tokens. In this work, we unify these seemingly distinct objectives through a minimalist intervention, namely CoPE: soft clipping lowfrequency components of RoPE. CoPE not only eliminates OOD outliers and refines semantic signals, but also prevents spectral leakage caused by hard clipping. Extensive experiments demonstrate that simply applying our soft clipping strategy to RoPE yields significant performance gains that scale up to 256k context length, validating our theoretical analysis and establishing CoPE as a new state-of-the-art for length generalization. Our code, data, and models are available at https://github.com/hrlics/CoPE.