OScaR: The Occam's Razor for Extreme KV Cache Quantization in LLMs and Beyond

arXiv:2605.19660 · cs.LG, cs.CL · Submitted 2026-05-19 · Read on arXiv

cs.LG, cs.CL

Submitted: 2026-05-19

Updated: 2026-10-14

Code: https://github.com/ZunhaiSu/OScaR-KV-Quant

Terminology

• Quantization — 7× in this paper · explained in 12 episodes
• KV cache — 2× in this paper · explained in 3 episodes
• Fidelity — 1× in this paper · explained in 2 episodes
• Token Norm Imbalance — 1× in this paper

Sources

• HadaCore: Tensor Core Accelerated Hadamard Transform Kernel
• gpt-oss-120b & gpt-oss-20b Model Card
• Systematic Outliers in Large Language Models
• Qwen3-VL Technical Report
• PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling
• Qwen3-Coder-Next Technical Report
• SKVQ: Sliding-window Key and Value Cache Quantization for Large Language Models
• GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers
• Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs
• Active-Dormant Attention Heads: Mechanistically Demystifying Extreme-Token Phenomena in LLMs
• PolarQuant: Quantizing KV Caches with Polar Transformation
• WorldSense: Evaluating Real-world Omnimodal Understanding for Multimodal LLMs
• IsoQuant: Hardware-Aligned SO(4) Isoclinic Rotations for LLM KV Cache Compression
• Massive Values in Self-Attention Modules are the Key to Contextual Knowledge Understanding
• LLaVA-OneVision: Easy Visual Task Transfer
• A Survey on Large Language Model Acceleration based on KV Cache Management
• Memory-Efficient Visual Autoregressive Modeling with Scale-Aware KV Cache Compression
• TWEO: Transformers Without Extreme Outliers Enables FP8 Training And Quantization For Dummies
• KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache
• A White Paper on Neural Network Quantization

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