# Cartridges: Lightweight and general-purpose long context representations via self-study

arXiv: https://arxiv.org/abs/2506.06266 · HTML: https://mlsanity.com/paper/2506.06266/

## Key facts

- Cited by 86 papers in the map (data through 14 Oct 2026).
- Placed in Prompt and context compression, one of 12 directions in Efficient LLM inference and serving.
- Builds on 104 of its 121 references that are in the map.
- First posted on arXiv 6 Jun 2025.

## Placed in

- Efficient LLM inference and serving / [Prompt and context compression](https://mlsanity.com/inference/prompt-and-context-compression/)

## Builds on (most cited first, 30 of 104)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [Qwen2.5 Technical Report](https://mlsanity.com/paper/2412.15115/) (2024)
- [Mamba: Linear-Time Sequence Modeling with Selective State Spaces](https://mlsanity.com/paper/2312.00752/) (2023)
- [SQuAD: 100,000+ Questions for Machine Comprehension of Text](https://mlsanity.com/paper/1606.05250/) (2016)
- [Prefix-Tuning: Optimizing Continuous Prompts for Generation](https://mlsanity.com/paper/2101.00190/) (2021)
- [The Power of Scale for Parameter-Efficient Prompt Tuning](https://mlsanity.com/paper/2104.08691/) (2021)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [Lost in the Middle: How Language Models Use Long Contexts](https://mlsanity.com/paper/2307.03172/) (2023)
- [Longformer: The Long-Document Transformer](https://mlsanity.com/paper/2004.05150/) (2020)
- [Efficient Streaming Language Models with Attention Sinks](https://mlsanity.com/paper/2309.17453/) (2023)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters](https://mlsanity.com/paper/2408.03314/) (2024)
- [Big Bird: Transformers for Longer Sequences](https://mlsanity.com/paper/2007.14062/) (2020)
- [Generating Long Sequences with Sparse Transformers](https://mlsanity.com/paper/1904.10509/) (2019)
- [MemGPT: Towards LLMs as Operating Systems](https://mlsanity.com/paper/2310.08560/) (2023)
- [GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints](https://mlsanity.com/paper/2305.13245/) (2023)
- [RULER: What's the Real Context Size of Your Long-Context Language Models?](https://mlsanity.com/paper/2404.06654/) (2024)
- [BitFit: Simple Parameter-efficient Fine-tuning for Transformer-based Masked Language-models](https://mlsanity.com/paper/2106.10199/) (2021)
- [SGLang: Efficient Execution of Structured Language Model Programs](https://mlsanity.com/paper/2312.07104/) (2023)
- [DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model](https://mlsanity.com/paper/2405.04434/) (2024)
- [SnapKV: LLM Knows What You are Looking for Before Generation](https://mlsanity.com/paper/2404.14469/) (2024)
- [H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models](https://mlsanity.com/paper/2306.14048/) (2023)
- [Sequence-Level Knowledge Distillation](https://mlsanity.com/paper/1606.07947/) (2016)
- [Fast Transformer Decoding: One Write-Head is All You Need](https://mlsanity.com/paper/1911.02150/) (2019)
- [Phi-4 Technical Report](https://mlsanity.com/paper/2412.08905/) (2024)

## Built on by (most cited first, 30 of 86)

- [Reinforcement Learning via Self-Distillation](https://mlsanity.com/paper/2601.20802/) (2026)
- [Self-Distillation Enables Continual Learning](https://mlsanity.com/paper/2601.19897/) (2026)
- [MIRIX: Multi-Agent Memory System for LLM-Based Agents](https://mlsanity.com/paper/2507.07957/) (2025)
- [Nested Learning: The Illusion of Deep Learning Architectures](https://mlsanity.com/paper/2512.24695/) (2025)
- [Cache-to-Cache: Direct Semantic Communication Between Large Language Models](https://mlsanity.com/paper/2510.03215/) (2025)
- [Expanding the Capabilities of Reinforcement Learning via Text Feedback](https://mlsanity.com/paper/2602.02482/) (2026)
- [Doc-to-LoRA: Learning to Instantly Internalize Contexts](https://mlsanity.com/paper/2602.15902/) (2026)
- [Aligning Language Models from User Interactions](https://mlsanity.com/paper/2603.12273/) (2026)
- [MemR3: Memory Retrieval via Reflective Reasoning for LLM Agents](https://mlsanity.com/paper/2512.20237/) (2025)
- [Continual Learning Bench: Evaluating Frontier AI Systems in Real-World Stateful Environments](https://mlsanity.com/paper/2606.05661/) (2026)
- [Pretraining with hierarchical memories: separating long-tail and common knowledge](https://mlsanity.com/paper/2510.02375/) (2025)
- [KV Cache Transform Coding for Compact Storage in LLM Inference](https://mlsanity.com/paper/2511.01815/) (2025)
- [Self-Compacting Language Model Agents](https://mlsanity.com/paper/2606.23525/) (2026)
- [EVICPRESS: Joint KV-Cache Compression and Eviction for Efficient LLM Serving](https://mlsanity.com/paper/2512.14946/) (2025)
- [Prompt-MII: Meta-Learning Instruction Induction for LLMs](https://mlsanity.com/paper/2510.16932/) (2025)
- [GradMem: Learning to Write Context into Memory with Test-Time Gradient Descent](https://mlsanity.com/paper/2603.13875/) (2026)
- [An Information Theoretic Perspective on Agentic System Design](https://mlsanity.com/paper/2512.21720/) (2025)
- [Dynamic Long Context Reasoning over Compressed Memory via End-to-End Reinforcement Learning](https://mlsanity.com/paper/2602.08382/) (2026)
- [Position: Modular Memory is the Key to Continual Learning Agents](https://mlsanity.com/paper/2603.01761/) (2026)
- [EpiCache: Episodic KV Cache Management for Long-Term Conversation on Resource-Constrained Environments](https://mlsanity.com/paper/2509.17396/) (2025)
- [Training Transformers for KV Cache Compressibility](https://mlsanity.com/paper/2605.05971/) (2026)
- [When Does Continual Learning Require Learning](https://mlsanity.com/paper/2607.07847/) (2026)
- [Cartridges at Scale: Training Modular KV Caches over Large Document Collections](https://mlsanity.com/paper/2606.04557/) (2026)
- [Do LLMs Benefit From Their Own Words?](https://mlsanity.com/paper/2602.24287/) (2026)
- [A Neuropsychologically Grounded Evaluation of LLM Cognitive Abilities](https://mlsanity.com/paper/2603.02540/) (2026)
- [Parallel Context Compaction for Long-Horizon LLM Agent Serving](https://mlsanity.com/paper/2605.23296/) (2026)
- [Nearly Optimal Attention Coresets](https://mlsanity.com/paper/2605.05602/) (2026)
- [MeMo: Memory as a Model](https://mlsanity.com/paper/2605.15156/) (2026)
- [Still: Amortized KV Cache Compaction in a Single Forward Pass](https://mlsanity.com/paper/2606.07878/) (2026)
- [End-to-End Context Compression at Scale](https://mlsanity.com/paper/2606.09659/) (2026)

## Abstract (arXiv)

Large language models are often used to answer queries grounded in large text corpora (e.g. codebases, legal documents, or chat histories) by placing the entire corpus in the context window and leveraging in-context learning (ICL). Although current models support contexts of 100K-1M tokens, this setup is costly to serve because the memory consumption of the KV cache scales with input length. We explore an alternative: training a smaller KV cache offline on each corpus. At inference time, we load this trained KV cache, which we call a Cartridge, and decode a response. Critically, the cost of training a Cartridge can be amortized across all the queries referencing the same corpus. However, we find that the naive approach of training the Cartridge with next-token prediction on the corpus is not competitive with ICL. Instead, we propose self-study, a training recipe in which we generate synthetic conversations about the corpus and train the Cartridge with a context-distillation objective. We find that Cartridges trained with self-study replicate the functionality of ICL, while being significantly cheaper to serve. On challenging long-context benchmarks, Cartridges trained with self-study match ICL performance while using 38.6x less memory and enabling 26.4x higher throughput. Self-study also extends the model's effective context length (e.g. from 128k to 484k tokens on MTOB) and surprisingly, leads to Cartridges that can be composed at inference time without retraining.