# MG-Cache: Multi-Granularity Caching for Accelerating Diffusion Transformer Inference

DevFeed: [MG-Cache: Multi-Granularity Caching for Accelerating Diffusion Transformer Inference](<https://devfeed.tech/articles/mg-cache-multi-granularity-caching-for-accelerating-diffusion-transformer-inference-125367.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11365>)

Author: Zesen Chen, Youxuan Xu, Shigang Li

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [diffusion-transformers](<https://devfeed.tech/topics/diffusion-transformers.md>), [Cache](<https://devfeed.tech/topics/cache.md>)

Tags: [acceleration](<https://devfeed.tech/tags/acceleration.md>), [adaptive](<https://devfeed.tech/tags/adaptive.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [attention](<https://devfeed.tech/tags/attention.md>), [cache](<https://devfeed.tech/tags/cache.md>), [caching](<https://devfeed.tech/tags/caching.md>), [caching-strategies](<https://devfeed.tech/tags/caching-strategies.md>), [control](<https://devfeed.tech/tags/control.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-pf](<https://devfeed.tech/tags/cs-pf.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [effectiveness](<https://devfeed.tech/tags/effectiveness.md>), [error](<https://devfeed.tech/tags/error.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [exploit](<https://devfeed.tech/tags/exploit.md>), [free](<https://devfeed.tech/tags/free.md>), [generative](<https://devfeed.tech/tags/generative.md>), [images](<https://devfeed.tech/tags/images.md>), [inference](<https://devfeed.tech/tags/inference.md>), [innovation](<https://devfeed.tech/tags/innovation.md>), [transformer](<https://devfeed.tech/tags/transformer.md>)

## AI overview

MG-Cache accelerates Diffusion Transformer inference without additional training by coordinating cache reuse across timesteps and Transformer layers. It first profiles timesteps offline to identify opportunities for reuse, then adaptively caches stable Attention and MLP submodules based on input stability and layer importance. The authors report experiments on mainstream open-source DiT models with significant speedups while retaining high fidelity to uncached outputs, including up to 4.18x acceleration on HunyuanVideo. The excerpt does not provide further experimental details or comparisons.

## Source excerpt

arXiv:2611.11365v1 Announce Type: new Abstract: Diffusion Transformer (DiT) models, while having become a powerful backbone for generative synthesis, suffer from high inference cost on generating high-resolution images. Diffusion caching has emerged as a promising acceleration technique to reduce inference cost. However, existing caching strategies fail to accurately identify the varying importance of different computational modules and rarely coordinate cache reusing across the timestep and layer granularities, making it difficult to control error accumulation. To address these limitations, we propose multi-granularity caching (MG-Cache) for DiTs, a novel and fully training-free approach to exploit computational redundancy both on timesteps and Transformer layers. Our core innovation contains a two-stage caching strategy: an offline timestep profiling analysis that identifies redundant timesteps for cache reuse, followed by an adaptive layer-wise caching strategy that combines input stability with importance of each Transformer layer to selectively cache stable Attention and MLP submodules. Extensive experiments on mainstream open-source DiT models validate the effectiveness of our approach, achieving significant acceleration while maintaining high fidelity to the uncached output, with up to a 4.18x acceleration on HunyuanVideo.