홈 › Papers

ParallelComp: Parallel Long-Context Compressor for Length Extrapolation

Efficiently handling long contexts is crucial for large language models (LLMs). While rotary position embeddings (RoPEs) enhance length generalization, effective length extrapolation remains challenging and often requires costly fine-tuning. In contrast, recent training-free approaches suffer from the attention sink phenomenon, leading to severe performance degradation. In this paper, we introduce ParallelComp, a novel training-free method for long-context extrapolation that extends LLMs' context length from 4K to 128K while maintaining high throughput and preserving perplexity, and integrates seamlessly with Flash Attention. Our analysis offers new insights into attention biases in parallel attention mechanisms and provides practical solutions to tackle these challenges. To mitigate the attention sink issue, we propose an attention calibration strategy that reduces biases, ensuring more stable long-range attention. Additionally, we introduce a chunk eviction strategy to efficiently manage ultra-long contexts on a single A100 80GB GPU. To further enhance efficiency, we propose a parallel KV cache eviction technique, which improves chunk throughput by 1.76x, thereby achieving a 23.50x acceleration in the prefilling stage with negligible performance loss due to attention calibration. Furthermore, ParallelComp achieves 91.17% of GPT-4's performance on long-context tasks using an 8B model trained on 8K-length context, outperforming powerful closed-source models such as Claude-2 and Kimi-Chat.

Code (0)

등록된 구현이 없습니다.

Tasks

Methods 이 논문이 사용한 방법론

Similar Papers 제목 키워드 기반