Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Suraj Ranganath, Vaishak Menon, Anish Patnaik

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/.

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

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/.

Paper Structure

This paper contains 19 sections, 10 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method Families and Benchmark Harness
  4. Problem Setting
  5. Benchmarks
  6. Method Families
  7. Evaluation Axes
  8. Experimental Setup
  9. Results
  10. Global Trade-offs Across 33 Methods
  11. Pareto and Frontier Analysis
  12. Qualitative Behavior
  13. Why Does Peak VRAM Sometimes Increase Despite Compression?
  14. Discussion
  15. Future Work
  16. ...and 4 more sections

Figures (10)

  • Figure 1: Global systems-quality landscape on MovieGen. FlowCache-inspired prune/soft-prune methods dominate the practical low-VRAM region, PRQ and QuaRot occupy the high-fidelity region, and spatially mixed methods collapse despite plausible motivation.
  • Figure 2: Global systems-quality landscape on StoryEval. The qualitative structure mirrors MovieGen: the practical winner remains in the FlowCache-inspired soft-prune region, while the highest-fidelity compressed methods sit at much higher runtime or peak-memory cost.
  • Figure 3: Pareto/frontier analysis on MovieGen. Methods that survive the quality-preserving compression frontier are not the same methods that survive the systems-efficiency frontier, which is why deployment winners and research-quality winners diverge.
  • Figure 4: Pareto/frontier analysis on StoryEval. The same separation persists under narrative-style rollout stability, reinforcing that the design-space conclusions are not artifacts of one benchmark.
  • Figure 5: Curated six-method qualitative comparisons for two MovieGen examples and one StoryEval example.
  • ...and 5 more figures