kv-cache-quantization

OSCAR is an offline spectral covariance-aware rotation method for 2-bit KV cache quantization that aligns quantization with attention covariance structures, achieving high accuracy and efficiency for long-context LLM serving.

This paper analyzes KV cache quantization schemes inspired by TurboQuant, using statistical inference and a new 6D error framework to evaluate quality measures like KL divergence and geometric error.