The Golden Path Hypothesis: Reusable Schedules in Diffusion Caching
cs.AI, cs.LG
Submitted: 2026-09-30
Updated: 2026-09-30
Code: https://github.com/nanguoyu/Golden-Path-Hypothesis
Sources
- ReCache: Learning Budget-Aware Caching Schedules for Diffusion Models via REINFORCE
- DiCache: Let Diffusion Model Determine Its Own Cache
- On the Trajectory Regularity of ODE-based Diffusion Sampling
- Forecast the Principal, Stabilize the Residual: Subspace-Aware Feature Caching for Efficient Diffusion Transformers
- $\Delta$-DiT: A Training-Free Acceleration Method Tailored for Diffusion Transformers
- OmniCache: A Trajectory-Oriented Global Perspective on Training-Free Cache Reuse for Diffusion Transformer Models
- SeaCache: Spectral-Evolution-Aware Cache for Accelerating Diffusion Models
- Denoising as Path Planning: Training-Free Acceleration of Diffusion Models with DPCache
- HiCache: A Plug-in Scaled-Hermite Upgrade for Taylor-Style Cache-then-Forecast Diffusion Acceleration
- MeanCache: From Instantaneous to Average Velocity for Accelerating Flow Matching Inference
- OTCache: Optimal Transport for Geometry-Aware Caching in Diffusion Models
- SenCache: Accelerating Diffusion Model Inference via Sensitivity-Aware Caching
- HunyuanVideo: A Systematic Framework For Large Video Generative Models
- Budget-Constrained Step-Level Diffusion Caching
- Flow Matching for Generative Modeling
- Timestep Embedding Tells: It's Time to Cache for Video Diffusion Model
- From Reusing to Forecasting: Accelerating Diffusion Models with TaylorSeers
- FORA: Fast-Forward Caching in Diffusion Transformer Acceleration
- Relational Feature Caching for Accelerating Diffusion Transformers
- Wan: Open and Advanced Large-Scale Video Generative Models
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems