mlsanityData to 7 Oct 2026

Beyond Fixed Formulas: Data-Driven Linear Predictor for Efficient Diffusion Models

arXiv 2604.26365 ↗First posted 29 Apr 2026cs.CV cs.LGcited by 2

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

To address the high sampling cost of Diffusion Transformers (DiTs), feature caching offers a training-free acceleration method. However, existing methods rely on hand-crafted forecasting formulas that fail under aggressive skipping. We propose L2P (Learnable Linear Predictor), a simple data-driven caching framework that replaces fixed coefficients with learnable per-timestep weights. Rapidly trained in ~20 seconds on a single GPU, L2P accurately reconstructs current features from past trajectories. L2P significantly outperforms existing baselines: it achieves a 4.55x FLOPs reduction and 4.15x latency speedup on FLUX.1-dev, and maintains high visual fidelity under up to 7.18x acceleration on Qwen-Image models, where prior methods show noticeable quality degradation. Our results show learning linear predictors is highly effective for efficient DiT inference. Code is available at https://github.com/Aredstone/L2P-Cache.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 31 of its 39 references that are in the map. First posted on arXiv 29 Apr 2026.

Builds on

31 of 39 references.

Closest ancestors · 20

Same direction, newest first

  1. 2025A Survey on Cache Methods in Diffusion Models: Toward Efficient Multi-Modal Generation22 Oct
  2. 2025FreqCa: Accelerating Diffusion Models via Frequency-Aware Caching9 Oct
  3. 2025SpeCa: Accelerating Diffusion Transformers with Speculative Feature Caching15 Sep
  4. 2025Forecast then Calibrate: Feature Caching as ODE for Efficient Diffusion Transformers22 Aug
  5. 2025Accelerating Diffusion Model Training under Minimal Budgets: A Condensation-Based Perspective8 Jul
  6. 2025From Reusing to Forecasting: Accelerating Diffusion Models with TaylorSeers10 Mar
  7. 2025Cached Adaptive Token Merging: Dynamic Token Reduction and Redundant Computation Elimination in Diffusion Model1 Jan
  8. 2024Token Pruning for Caching Better: 9 Times Acceleration on Stable Diffusion for Free31 Dec
  9. 2024Rethinking Token-wise Feature Caching: Accelerating Diffusion Transformers with Dual Feature Caching25 Dec
  10. 2024DiP-GO: A Diffusion Pruner via Few-step Gradient Optimization22 Oct
  11. 2024Accelerating Diffusion Transformers with Token-wise Feature Caching5 Oct
  12. 2024HarmoniCa: Harmonizing Training and Inference for Better Feature Caching in Diffusion Transformer Acceleration2 Oct
8 more
  1. 2024Learning-to-Cache: Accelerating Diffusion Transformer via Layer Caching3 Jun
  2. 2024Delta-DiT: A Training-Free Acceleration Method Tailored for Diffusion Transformers3 Jun
  3. 2023DeepCache: Accelerating Diffusion Models for Free1 Dec
  4. 2023Consistency Models2 Mar
  5. 2022DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models2 Nov
  6. 2022DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps2 Jun
  7. 2020Denoising Diffusion Implicit Models6 Oct
  8. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 11

  1. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,604
  2. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,029
  3. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,550
  4. 2023Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasetscited by 2,526
  5. 2024CogVideoX: Text-to-Video Diffusion Models with An Expert Transformercited by 2,198
  6. 2024HunyuanVideo: A Systematic Framework For Large Video Generative Modelscited by 1,597
  7. 2025Qwen-Image Technical Reportcited by 973
  8. 2024Timestep Embedding Tells: It's Time to Cache for Video Diffusion Modelcited by 245
  9. 2024DiTFastAttn: Attention Compression for Diffusion Transformer Modelscited by 109
  10. 2024Adaptive Caching for Faster Video Generation with Diffusion Transformerscited by 100
  11. 2024PeRFlow: Piecewise Rectified Flow as Universal Plug-and-Play Acceleratorcited by 95

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026The Golden Path Hypothesis: Reusable Schedules in Diffusion Cachingcited by 0
  2. 2026Koopman Observers for Diffusion Acceleration: Correcting Feature Forecasts with Shallow Measurementscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Forecast the Principal, Stabilize the Residual: Subspace-Aware Feature Caching for Efficient Diffusion Transformers
  2. 2026LESA: Learnable Stage-Aware Predictors for Diffusion Model Acceleration
  3. 2026BeCARE: Budgeted Cache Refresh for Diffusion Transformer Acceleration
  4. 2025Forecasting When to Forecast: Accelerating Diffusion Models with Confidence-Gated Taylor
  5. 2026Predict to Skip: Linear Multistep Feature Forecasting for Efficient Diffusion Transformers
  6. 2026LearniBridge: Learnable Calibration of Feature Caching for Diffusion Models Acceleration
  7. 2025HiCache: A Plug-in Scaled-Hermite Upgrade for Taylor-Style Cache-then-Forecast Diffusion Acceleration
  8. 2026Relational Feature Caching for Accelerating Diffusion Transformers