InfiniPot: Infinite Context Processing on Memory-Constrained LLMs

Minsoo Kim, Kyuhong Shim, Jungwook Choi, Simyung Chang


Abstract
Handling long input contexts remains a significant challenge for Large Language Models (LLMs), particularly in resource-constrained environments such as mobile devices. Our work aims to address this limitation by introducing InfiniPot, a novel KV cache control framework designed to enable pre-trained LLMs to manage extensive sequences within fixed memory constraints efficiently, without requiring additional training. InfiniPot leverages Continual Context Distillation (CCD), an iterative process that compresses and retains essential information through novel importance metrics, effectively maintaining critical data even without access to future context. Our comprehensive evaluations indicate that InfiniPot significantly outperforms models trained for long contexts in various NLP tasks, establishing its efficacy and versatility. This work represents a substantial advancement toward making LLMs applicable to a broader range of real-world scenarios.
Anthology ID:
2024.emnlp-main.897
Volume:
Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing
Month:
November
Year:
2024
Address:
Miami, Florida, USA
Editors:
Yaser Al-Onaizan, Mohit Bansal, Yun-Nung Chen
Venue:
EMNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
16046–16060
Language:
URL:
https://aclanthology.org/2024.emnlp-main.897
DOI:
10.18653/v1/2024.emnlp-main.897
Bibkey:
Cite (ACL):
Minsoo Kim, Kyuhong Shim, Jungwook Choi, and Simyung Chang. 2024. InfiniPot: Infinite Context Processing on Memory-Constrained LLMs. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 16046–16060, Miami, Florida, USA. Association for Computational Linguistics.
Cite (Informal):
InfiniPot: Infinite Context Processing on Memory-Constrained LLMs (Kim et al., EMNLP 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.emnlp-main.897.pdf