Finch: Prompt-guided Key-Value Cache Compression

TL;DR

Finch is a novel method that compresses long input texts for large language models by selectively caching relevant key-value pairs, enabling processing of longer inputs with high compression ratios without fine-tuning.

Contribution

Finch introduces a prompt-guided key-value cache compression technique that significantly reduces memory usage while maintaining semantic integrity for long inputs.

Findings

Achieves up to 93x compression of input texts

Enables models to process longer contexts within fixed GPU memory

Preserves semantic content despite high compression

Abstract

Recent large language model applications, such as Retrieval-Augmented Generation and chatbots, have led to an increased need to process longer input contexts. However, this requirement is hampered by inherent limitations. Architecturally, models are constrained by a context window defined during training. Additionally, processing extensive texts requires substantial GPU memory. We propose a novel approach, Finch, to compress the input context by leveraging the pre-trained model weights of the self-attention. Given a prompt and a long text, Finch iteratively identifies the most relevant Key (K) and Value (V) pairs over chunks of the text conditioned on the prompt. Only such pairs are stored in the KV cache, which, within the space constrained by the context window, ultimately contains a compressed version of the long text. Our proposal enables models to consume large inputs even with high compression (up to 93x) while preserving semantic integrity without the need for fine-tuning.