Knowledge Capsules: Structured Nonparametric Memory Units for LLMs

TL;DR

The paper introduces Knowledge Capsules, a structured memory system for LLMs that enhances knowledge integration stability and accuracy without retraining, outperforming retrieval-augmented methods in QA tasks.

Contribution

It presents a novel External Key Value Injection framework that directly incorporates structured knowledge into attention mechanisms, improving over existing retrieval-based approaches.

Findings

Outperforms RAG and GraphRAG on multiple QA benchmarks.

Provides improved stability and accuracy in long context and multi-hop reasoning.

Requires no parameter updates during knowledge integration.

Abstract

Large language models (LLMs) encode knowledge in parametric weights, making it costly to update or extend without retraining. Retrieval-augmented generation (RAG) mitigates this limitation by appending retrieved text to the input, but operates purely through context expansion, where external knowledge competes as tokens within the attention mechanism. As a result, its influence is indirect and often unstable, particularly in long context and multi hop reasoning scenarios. We propose Knowledge Capsules, structured nonparametric memory units that represent normalized relational knowledge and can be constructed directly from document corpora using a frozen base model. Instead of injecting knowledge as text, we introduce an External Key Value Injection (KVI) framework that compiles capsules into attention-compatible key value representations, enabling external knowledge to directly participate in the model's attention computation. By shifting knowledge integration from context-level augmentation to memory level interaction, the proposed framework consistently outperforms RAG and GraphRAG across multiple QA benchmarks, with improved stability and accuracy in long context and multi hop reasoning, while requiring no parameter updates.