Forcing-KV: Hybrid KV Cache Compression for Efficient Autoregressive Video Diffusion Models

Hugging Face Daily Papers Papers
kv-cache-compression video-diffusion autoregressive attention-mechanism efficiency deep-learning

Summary

This paper introduces Forcing-KV, a hybrid KV cache compression strategy for autoregressive video diffusion models that separates attention heads into static and dynamic categories, achieving up to 2.82x speedup at 1080P resolution while maintaining output quality.

Autoregressive (AR) video diffusion models adopt a streaming generation framework, enabling long-horizon video generation with real-time responsiveness, as exemplified by the Self Forcing training paradigm. However, existing AR video diffusion models still suffer from significant attention complexity and severe memory overhead due to the redundant key-value (KV) caches across historical frames, which limits scalability. In this paper, we tackle this challenge by introducing KV cache compression into autoregressive video diffusion. We observe that attention heads in mainstream AR diffusion models exhibit markedly distinct attention patterns and functional roles that remain stable across samples and denoising steps. Building on our empirical study of head-wise functional specialization, we divide the attention heads into two categories: static heads, which focus on transitions across autoregressive chunks and intra-frame fidelity, and dynamic heads, which govern inter-frame motion and consistency. We then propose Forcing-KV, a hybrid KV cache compression strategy that performs structured static pruning for static heads and dynamic pruning based on segment-wise similarity for dynamic heads. While maintaining output quality, our method achieves a generation speed of over 29 frames per second on a single NVIDIA H200 GPU along with 30% cache memory reduction, delivering up to 1.35x and 1.50x speedups on LongLive and Self Forcing at 480P resolution, and further scaling to 2.82x speedup at 1080P resolution. Code and demo videos are provided at https://zju-jiyicheng.github.io/Forcing-KV-Page.
Original Article
View Cached Full Text

Cached at: 05/15/26, 08:24 AM

Paper page - Forcing-KV: Hybrid KV Cache Compression for Efficient Autoregressive Video Diffusion Models

Source: https://huggingface.co/papers/2605.09681

Abstract

Autoregressive video diffusion models face scalability issues due to high attention complexity and memory overhead from redundant key-value caches, which are addressed through a hybrid compression strategy that separates attention heads into static and dynamic categories for optimized caching.

Autoregressive (AR) video diffusion models adopt astreaming generation framework, enabling long-horizon video generation with real-time responsiveness, as exemplified by theSelf Forcing training paradigm. However, existing AR video diffusion models still suffer from significantattention complexityand severememory overheaddue to the redundant key-value (KV) caches across historical frames, which limits scalability. In this paper, we tackle this challenge by introducingKV cache compressioninto autoregressive video diffusion. We observe thatattention headsin mainstream AR diffusion models exhibit markedly distinct attention patterns and functional roles that remain stable across samples and denoising steps. Building on our empirical study of head-wise functional specialization, we divide theattention headsinto two categories:static heads, which focus on transitions across autoregressive chunks and intra-frame fidelity, anddynamic heads, which govern inter-frame motion and consistency. We then propose Forcing-KV, a hybridKV cache compressionstrategy that performsstructured static pruningforstatic headsanddynamic pruningbased onsegment-wise similarityfordynamic heads. While maintaining output quality, our method achieves a generation speed of over 29 frames per second on a single NVIDIA H200 GPU along with 30% cache memory reduction, delivering up to 1.35x and 1.50x speedups on LongLive and Self Forcing at 480P resolution, and further scaling to 2.82x speedup at 1080P resolution. Code and demo videos are provided at https://zju-jiyicheng.github.io/Forcing-KV-Page.

View arXiv pageView PDFProject pageGitHub66Add to collection

Get this paper in your agent:

hf papers read 2605\.09681

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.09681 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.09681 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.09681 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles