Learnable Linear Predictor (L2P)
Learnable Linear Predictor (L2P)
Auto-generated stub. Edit this file to add more details.
A data-driven feature caching framework for diffusion models that replaces fixed forecasting formulas with learnable per-timestep weights.
Why It Matters
It provides a systematic, data-driven alternative to rigid, heuristic-based feature caching in diffusion models, enabling more aggressive skip-sampling without quality degradation.
Evidence
We propose L2P (Learnable Linear Predictor), a simple data-driven caching framework that replaces fixed coefficients with learnable per-timestep weights.
Related Papers
Metadata & Links
- created_at
- 2026-05-02T05:50:03Z