Cartridges++: KV Cache Compression without Off-Context Derailment

arXiv:2609.35621 · cs.LG · Submitted 2026-09-28 · Read on arXiv

cs.LG

Submitted: 2026-09-28

Updated: 2026-09-28

Code: https://github.com/HazyResearch/cartridges

Terminology

• KV cache — 1× in this paper · explained in 3 episodes

Sources

• gpt-oss-120b & gpt-oss-20b Model Card
• Does Accuracy Equal Evidence? Reasoning Faithfulness under KV Cache Compression
• Doc-to-LoRA: Learning to Instantly Internalize Contexts
• KV-Distill: Nearly Lossless Learnable Context Compression for LLMs
• DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
• DeepSeek-V3 Technical Report
• Doc-to-Atom: Learning to Compile and Compose Memory Atoms
• Learned Structure in Cartridges: Keys as Shareable Routers in Self-Studied Representations
• Training Transformers for KV Cache Compressibility
• Cartridges at Scale: Training Modular KV Caches over Large Document Collections
• When Does Continual Learning Require Learning
• Latent Context Compilation: Distilling Long Context into Compact Portable Memory
• Semantic Integrity Matters: Benchmarking and Preserving High-Density Reasoning in KV Cache Compression
• Nectar: Neural Estimation of Cached-Token Attention via Regression
• Learning to Evict from Key-Value Cache
• SGPT: GPT Sentence Embeddings for Semantic Search
• Long Context In-Context Compression by Getting to the Gist of Gisting
• What It Costs to Compose, Rebuild, and Correct Precomputed Memory
• Progressive Cramming: Reliable Token Compression and What It Reveals
• Gemma 3 Technical Report

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks