KV Cache Quantization for Self-Forcing Video Generation: A 33-Method Empirical Study

• URL: http://arxiv.org/abs/2603.27469v1
• Date: Sun, 29 Mar 2026 01:35:16 GMT
• Title: KV Cache Quantization for Self-Forcing Video Generation: A 33-Method Empirical Study
• Abstract summary: We present a comprehensive study of KV-cache compression for self-forcing video generation on a Wan2.1-based Self-Forcing stack.<n>Our study covers 33 quantization and cache-policy variants, 610 prompt-level observations, and 63 benchmark-level summaries.<n>We jointly evaluate peak VRAM, runtime, realized compression ratio, VBench imaging quality, BF16-referenced fidelity (SSIM, LPIPS, PSNR), and terminal drift.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Self-forcing video generation extends a short-horizon video model to longer rollouts by repeatedly feeding generated content back in as context. This scaling path immediately exposes a systems bottleneck: the key-value (KV) cache grows with rollout length, so longer videos require not only better generation quality but also substantially better memory behavior. We present a comprehensive empirical study of KV-cache compression for self-forcing video generation on a Wan2.1-based Self-Forcing stack. Our study covers 33 quantization and cache-policy variants, 610 prompt-level observations, and 63 benchmark-level summaries across two evaluation settings: MovieGen for single-shot 10-second generation and StoryEval for longer narrative-style stability. We jointly evaluate peak VRAM, runtime, realized compression ratio, VBench imaging quality, BF16-referenced fidelity (SSIM, LPIPS, PSNR), and terminal drift. Three findings are robust. First, the strongest practical operating region is a FlowCache-inspired soft-prune INT4 adaptation, which reaches 5.42-5.49x compression while reducing peak VRAM from 19.28 GB to about 11.7 GB with only modest runtime overhead. Second, the highest-fidelity compressed methods, especially PRQ_INT4 and QUAROT_KV_INT4, are not the best deployment choices because they preserve quality at severe runtime or memory cost. Third, nominal compression alone is not sufficient: several methods shrink KV storage but still exceed BF16 peak VRAM because the current integration reconstructs or retains large BF16 buffers during attention and refresh stages. The result is a benchmark harness, analysis workflow, and empirical map of which KV-cache ideas are practical today and which are promising research directions for better memory integration. Code, data products, and the presentation dashboard are available at https://github.com/suraj-ranganath/kv-quant-longhorizon/.

Related papers

• MuKV: Multi-Grained KV Cache Compression for Long Streaming Video Question-Answering [75.0394545769057]
  KV-caching stores the Key-Value of the historical tokens via LLM prefill.<n>MuKV is a method that features a multi-grained KV cache compression module and a semi-hierarchical retrieval approach.<n> Experiments on long-streaming VideoQA benchmarks show that MuKV significantly improves answer accuracy, without sacrificing memory and online QA efficiency.
  arXiv  Detail & Related papers  (2026-05-21T10:13:03Z)
• PackForcing: Short Video Training Suffices for Long Video Sampling and Long Context Inference [46.18482046594169]
  PackForcing generates coherent 2-minute, 832x480 videos at 16 FPS on a single H200 GPU.<n>It achieves a bounded KV cache of just 4 GB and enables a remarkable 24x temporal (5s to 120s) operating effectively either zero-shot or trained on merely 5-second clips.
  arXiv  Detail & Related papers  (2026-03-26T17:59:05Z)
• Quant VideoGen: Auto-Regressive Long Video Generation via 2-Bit KV-Cache Quantization [83.406036390582]
  Quant VideoGen (QVG) is a training free KV cache quantization framework for autoregressive video diffusion models.<n>It reduces KV memory by up to 7.0 times with less than 4% end to end latency overhead.<n>It consistently outperforms existing baselines in generation quality.
  arXiv  Detail & Related papers  (2026-02-03T00:54:32Z)
• Past- and Future-Informed KV Cache Policy with Salience Estimation in Autoregressive Video Diffusion [53.14908419375226]
  existing approaches generally rely on KV Cache policies, which ignore differences in token importance in long-term video generation.<n>We propose a novel Past- and Future-Informed KV Cache Policy (PaFu-KV)<n>Specifically, PaFu-KV introduces a lightweight Salience Estimation Head distilled from a bidirectional cache teacher to estimate salience scores.<n>This policy yields a better quality-efficiency trade-off by shrinking KV cache capacity and reducing memory footprint at inference time.
  arXiv  Detail & Related papers  (2026-01-29T15:55:29Z)
• PackCache: A Training-Free Acceleration Method for Unified Autoregressive Video Generation via Compact KV-Cache [61.57938553036056]
  We introduce PackCache, a training-free KV-cache management method that compacts the KV cache through three coordinated mechanisms.<n>In terms of efficiency, PackCache accelerates end-to-end generation by 1.7-2.2x on 48-frame long sequences.
  arXiv  Detail & Related papers  (2026-01-07T19:51:06Z)
• StreamKV: Streaming Video Question-Answering with Segment-based KV Cache Retrieval and Compression [95.59657871147846]
  We propose textbfStreamKV, a framework that seamlessly equips Video-LLMs with advanced KV cache retrieval and compression.<n>Experiments on public StreamingVQA benchmarks demonstrate that StreamKV significantly outperforms existing Online Video-LLMs.
  arXiv  Detail & Related papers  (2025-11-10T16:25:03Z)
• HCAttention: Extreme KV Cache Compression via Heterogeneous Attention Computing for LLMs [13.013668526921778]
  Existing KV cache compression methods exhibit noticeable performance degradation when memory is reduced by more than 85%.<n>We propose HCAttention, a heterogeneous attention framework that integrates key quantization, value offloading, and dynamic KV eviction.<n> Experimental results on the LongBench benchmark demonstrate that our approach preserves the accuracy of full-attention model while shrinking the KV cache memory footprint to 25% of its original size.
  arXiv  Detail & Related papers  (2025-07-26T06:43:14Z)
• SpeCache: Speculative Key-Value Caching for Efficient Generation of LLMs [44.41154292836592]
  We propose SpeCache, which offloads the complete KV cache and dynamically fetches KV pairs back in each decoding step.<n> Experiments on LongBench and Needle-in-a-Haystack benchmarks verify that SpeCache effectively reduces VRAM usage.
  arXiv  Detail & Related papers  (2025-03-20T14:01:56Z)
• KVSharer: Efficient Inference via Layer-Wise Dissimilar KV Cache Sharing [58.29726147780976]
  We propose a plug-and-play method called textit KVSharer, which shares the KV cache between layers to achieve layer-wise compression. Experiments show that textit KVSharer can reduce KV cache computation by 30%, thereby lowering memory consumption. We verify that textit KVSharer is compatible with existing intra-layer KV cache compression methods, and combining both can further save memory.
  arXiv  Detail & Related papers  (2024-10-24T08:06:41Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.