feature-caching

[ICML 2025] This is the official PyTorch implementation of "🎵 HarmoniCa: Harmonizing Training and Inference for Better Feature Caching in Diffusion Transformer Acceleration".

[ICCV2025]LazyMAR: Accelerating Masked Autoregressive Models via Feature Caching

[ICLR 2026] Code for Evolutionary Caching to Accelerate Your Off-the-Shelf Diffusion Model

Drop-in TaylorSeer/HiCache basis upgrade — training-free diffusion acceleration via a Dynamic Mode Decomposition (Prony) exponential feature-forecast basis. Not the SGLang KV-cache HiCache.

Deploying a Multimodal Recommender System on Kubernetes featuring Cold Start handling, Bloom Filters, and Feature Caching.