English

Quantized Keys Steal Attention: Bias Correction for KV-Cache Compression in Video Diffusion

Machine Learning 2026-05-27 v1 Artificial Intelligence Computer Vision and Pattern Recognition Graphics Image and Video Processing

Abstract

Chunk-wise autoregressive video diffusion models rely on a KV cache of previously generated chunks to avoid redundant computation, but this cache quickly becomes a memory bottleneck as videos grow longer. Methods that quantize the KV cache to low bitwidths reduce memory pressure but degrade video quality. We show that a key driver of this degradation is a systematic bias in attention weights: due to the convexity of the exponential in softmax attention, quantization noise inflates the contribution of cached keys, a phenomenon we call the Jensen bias. This effect causes quantized keys to steal attention mass from the unquantized current chunk. We derive a per-attention-score correction that removes this bias in expectation, computed on the fly from the quantization step sizes of the cached keys and the query norm. Using a second-order Taylor approximation, the additional computational overhead is negligible, and no additional memory is needed alongside the cache. Evaluated on MAGI-1, SkyReels-V2, and HY-WorldPlay at INT2 quantization, our correction recovers most of the quality lost to aggressive quantization, reaching near-BF16 video quality, and can outperform INT4 quantization while using 50% less memory.

Keywords

Cite

@article{arxiv.2605.26266,
  title  = {Quantized Keys Steal Attention: Bias Correction for KV-Cache Compression in Video Diffusion},
  author = {Tuna Tuncer and Felix Becker and Thomas Pfeil},
  journal= {arXiv preprint arXiv:2605.26266},
  year   = {2026}
}

Comments

Variants of this manuscript were accepted to the ICML 2026 workshops SCALE and F2S