📒A curated list of Awesome Diffusion Inference Papers with codes. For Awesome LLM Inference, please check 📖Awesome-LLM-Inference 
for more details.
📖 News 🔥🔥
- [2026/03] Cache-DiT 🎉v1.3.0 release is ready, the major updates including: Ring Attention w/ batched P2P, USP (Hybrid Ring and Ulysses), Hybrid 2D and 3D Parallelism (💥USP + TP), VAE-P Comm overhead reduce.
🤖Contents
©️Citations
@misc{Awesome-DiT-Inference@2024,
title={Awesome-DiT-Inference: A small curated list of Awesome Diffusion Inference.},
url={https://github.com/xlite-dev/Awesome-DiT-Inference},
note={Open-source software available at https://github.com/xlite-dev/Awesome-DiT-Inference},
author={xlite-dev},
year={2024}
}
📙 Sampling
| Date | Title | Paper | Code | Recom |
|---|---|---|---|---|
| 2020.06 | 🔥[DDPM] Denoising Diffusion Probabilistic Models(@UC Berkeley) | [pdf] | [diffusion]  |
⭐️⭐️ |
| 2020.10 | 🔥[DDIM] DENOISING DIFFUSION IMPLICIT MODELS(@cs.stanford.edu) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2022.02 | 🔥[PNDM] PSEUDO NUMERICAL METHODS FOR DIFFUSION MODELS ON MANIFOLDS(@) | [pdf] | [PNDM]  |
⭐️⭐️ |
| 2022.02 | 🔥[DPM-Solver] DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps(@Cheng Lu) | [pdf] | [dpm-solver]  |
⭐️⭐️ |
| 2022.11 | 🔥[DPM-Solver++] DPM-SOLVER++: FAST SOLVER FOR GUIDED SAMPLING OF DIFFUSION PROBABILISTIC MODELS(@Cheng Lu) | [pdf] | [dpm-solver] |
⭐️⭐️ |
| 2023.10 | 🔥[DPM-Solver-v3] DPM-Solver-v3: Improved Diffusion ODE Solver with Empirical Model Statistics(@Kaiwen Zheng) | [pdf] | [DPM-Solver-v3]  |
⭐️⭐️ |
| 2023.11 | 🔥[Parallel Sampling] Parallel Sampling of Diffusion Models(@Stanford University) | [pdf] | [paradigms]  |
⭐️⭐️ |
| 2023.11 | 🔥[SAMPLER SCHEDULER] SAMPLER SCHEDULER FOR DIFFUSION MODELS(@sysu) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.02 | 🔥[Parallel Sampling] Accelerating Parallel Sampling of Diffusion Models(@Zhiwei Tang) | [pdf] | [ParaTAA-Diffusion]  |
⭐️⭐️ |
| 2024.01 | 🔥[YONOS] You Only Need One Step: Fast Super-Resolution with Stable Diffusion via Scale Distillation(@Samsung AI) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.01 | 🔥[S^2-DM] S^2-DMs: Skip-Step Diffusion Models(@Yixuan Wang) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.08 | 🔥[StepSaver] StepSaver: Predicting Minimum Denoising Steps for Diffusion Model Image Generation(@intel) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.09 | 🔥[DC-Solver] DC-Solver: Improving Predictor-Corrector Diffusion Sampler via Dynamic Compensation(@Tsinghua University) | [pdf] | [DC-Solver]  |
⭐️⭐️ |
📙 Caching
- UNet Based (DeepCache)
- DiT Based (Fast-Forward Caching)
| Date | Title | Paper | Code | Recom |
|---|---|---|---|---|
| 2023.05 | 🔥🔥[Cache-Enabled Sparse Diffusion] Accelerating Text-to-Image Editing via Cache-Enabled Sparse Diffusion Inference(@pku.edu.cn etc) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2023.12 | 🔥🔥[DeepCache] DeepCache: Accelerating Diffusion Models for Free(@nus.edu) | [pdf] | [DeepCache]  |
⭐️⭐️ |
| 2023.12 | 🔥🔥[Block Caching] Cache Me if You Can: Accelerating Diffusion Models through Block Caching(@Meta GenAI etc) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2023.12 | 🔥🔥[Approximate Caching] Approximate Caching for Efficiently Serving Diffusion Models(@Adobe) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.06 | 🔥🔥[Layer Caching] Learning-to-Cache: Accelerating Diffusion Transformer via Layer Caching(@nus.edu) | [pdf] | [learning-to-cache]  |
⭐️⭐️ |
| 2024.07 | 🔥[ElasticCache-LVLM] Efficient Inference of Vision Instruction-Following Models with Elastic Cache(@Tsinghua University etc) | [pdf] | [ElasticCache]  |
⭐️ |
| 2024.07 | 🔥🔥[Fast-Forward Caching(DiT)] FORA: Fast-Forward Caching in Diffusion Transformer Acceleration(@microsoft.com etc) | [pdf] | [FORA]  |
⭐️⭐️ |
| 2024.07 | 🔥🔥[Faster I2V Generation] Faster Image2Video Generation: A Closer Look at CLIP Image Embedding’s Impact on Spatio-Temporal Cross-Attentions(@Ashkan Taghipour etc) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.04 | 🔥🔥[T-GATE V1] Cross-Attention Makes Inference Cumbersome in Text-to-Image Diffusion Models(@Wentian Zhang etc) | [pdf] | [T-GATE]  |
⭐️⭐️ |
| 2024.04 | 🔥🔥[T-GATE V2] Faster Diffusion via Temporal Attention Decomposition(@Haozhe Liu etc) | [pdf] | [T-GATE] |
⭐️⭐️ |
| 2024.06 | 🔥🔥[DiTFastAttn] DiTFastAttn: Attention Compression for Diffusion Transformer Models(@Zhihang Yuan etc) | [pdf] | [DiTFastAttn]  |
⭐️⭐️ |
| 2024.06 | 🔥🔥[∆-DiT] ∆-DiT: A Training-Free Acceleration Method Tailored for Diffusion Transformers(@Fudan University) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.09 | 🔥🔥[TokenCache] Token Caching for Diffusion Transformer Acceleration(@Institute of Automation, Chinese Academy of Sciences) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.11 | 🔥🔥[AdaCache] Adaptive Caching for Faster Video Generation with Diffusion Transformers(@Meta) | [pdf] | [AdaCache]  |
⭐️⭐️ |
| 2024.11 | 🔥🔥[TeaCache] Timestep Embedding Tells: It’s Time to Cache for Video Diffusion Model(@Alibaba) | [pdf] | [TeaCache]  |
⭐️⭐️ |
| 2024.11 | 🔥🔥[LazyDiT] LazyDiT: Lazy Learning for the Acceleration of Diffusion Transformers(@Adobe Research) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.11 | 🔥🔥[Ca2-VDM] Ca2-VDM: Efficient Autoregressive Video Diffusion Model with Causal Generation and Cache Sharing(@ZJU) | [pdf] | [CausalCache-VDM]  |
⭐️⭐️ |
| 2024.11 | 🔥🔥[SmoothCache] SmoothCache: A Universal Inference Acceleration Technique for Diffusion Transformers(@Roblox) | [pdf] | [SmoothCache]  |
⭐️⭐️ |
| 2024.10 | 🔥🔥[FasterCache] FASTERCACHE: TRAINING-FREE VIDEO DIFFUSION MODEL ACCELERATION WITH HIGH QUALITY(@S-Lab) | [pdf] | [FasterCache]  |
⭐️⭐️ |
| 2024.10 | 🔥🔥[ToCa] ToCa: Accelerating Diffusion Transformers with Token-wise Feature Caching(@SJTU) | [pdf] | [ToCa]  |
⭐️⭐️ |
| 2024.11 | 🔥🔥[SkipCache] Accelerating Vision Diffusion Transformers with Skip Branches(@SJTU) | [pdf] | [Skip-DiT]  |
⭐️⭐️ |
| 2024.12 | 🔥🔥[DuCa] Accelerating Diffusion Transformers with Dual Feature Caching(@SJTU) | [pdf] | [DuCa]  |
⭐️⭐️ |
| 2025.01 | 🔥🔥[FBCache] Fastest HunyuanVideo Inference with Context Parallelism and First Block Cache on NVIDIA L20 GPUs(@chengzeyi) | [docs] | [ParaAttention]  |
⭐️⭐️ |
| 2025.01 | 🔥🔥[FlexCache] FlexCache: Flexible Approximate Cache System for Video Diffusion(@University of Waterloo) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2025.01 | 🔥🔥[Token Pruning] Token Pruning for Caching Better: 9× Acceleration on Stable Diffusion for Free(@SJTU) | [pdf] | [DaTo]  |
⭐️⭐️ |
| 2025.04 | 🔥🔥[AB-Cache] AB-Cache: Training-Free Acceleration of Diffusion Models via Adams-Bashforth Cached Feature Reuse(@USTC) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2025.03 | 🔥🔥[DiTFastAttnV2] DiTFastAttnV2: Head-wise Attention Compression for Multi-Modality Diffusion Transformers(@Infinigence AI) | [pdf] | [DiTFastAttn] |
⭐️⭐️ |
| 2025.03 | 🔥🔥[TaylorSeers] From Reusing to Forecasting: Accelerating Diffusion Models with TaylorSeers(@SJTU) | [pdf] | [TaylorSeer]  |
⭐️⭐️ |
| 2025.04 | 🔥🔥[Increment-Calibrated Cache] Accelerating Diffusion Transformer via Increment-Calibrated Caching with Channel-Aware Singular Value Decomposition(@PKU) | [pdf] | [icc]  |
⭐️⭐️ |
| 2025.05 | 🔥🔥[FastCache] FastCache: Fast Caching for Diffusion Transformer Through Learnable Linear Approximation(@yale) | [pdf] | [FastCache-xDiT]  |
⭐️⭐️ |
| 2025.06 | 🔥🔥[DBCache] DBCache: Dual Block Caching for Diffusion Transformers(@DefTruth, @vipshop, etc) | [docs] | [cache-dit]  |
⭐️⭐️ |
| 2025.06 | 🔥🔥[DBPrune] DBPrune: Dynamic Block Prune with Residual Caching(@DefTruth, @vipshop, etc) | [docs] | [cache-dit] |
⭐️⭐️ |
| 2025.06 | 🔥🔥[BACache] Block-wise Adaptive Caching for Accelerating Diffusion Policy(@THU) | [pdf] | ⚠️ | ⭐️⭐️ |
📙 Parallelism
- UNet Based: Displaced Patch parallelism (DistriFusion)
- DiT Based: Displaced Patch parallelism (PipeFusion)
| Date | Title | Paper | Code | Recom |
|---|---|---|---|---|
| 2024.02 | 🔥🔥[DistriFusion] DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Models(@MIT etc) | [pdf] | [distrifuser]  |
⭐️⭐️ |
| 2024.05 | 🔥🔥[PipeFusion] PipeFusion: Displaced Patch Pipeline Parallelism for Inference of Diffusion Transformer Models(@Tencent etc) | [pdf] | [xDiT]  |
⭐️⭐️ |
| 2024.06 | 🔥🔥[AsyncDiff] AsyncDiff: Parallelizing Diffusion Models by Asynchronous Denoising(@nus.edu) | [pdf] | [AsyncDiff]  |
⭐️⭐️ |
| 2024.05 | 🔥🔥[TensorRT-LLM SDXL] SDXL Distributed Inference with TensorRT-LLM and synchronous comm(@Zars19) | [pdf] | [SDXL-TensorRT-LLM]  |
⭐️⭐️ |
| 2024.06 | 🔥🔥[Clip Parallelism] Video-Infinity: Distributed Long Video Generation(@nus.edu) | [pdf] | [Video-Infinity]  |
⭐️⭐️ |
| 2024.05 | 🔥🔥[FIFO-Diffusion] FIFO-Diffusion: Generating Infinite Videos from Text without Training(@Seoul National University) | [pdf] | [FIFO-Diffusion]  |
⭐️⭐️ |
| 2025.01 | 🔥🔥[ParaAttention] Context parallel attention that accelerates DiT model inference with dynamic caching(@chengzeyi) | [docs] | [ParaAttention] |
⭐️⭐️ |
| 2026.01 | 🔥🔥[Cache-DiT] 🤗A PyTorch-native Inference Engine with Hybrid Cache Acceleration and Parallelism for DiTs. | [docs] | [cache-dit] |
⭐️⭐️ |
📙 Quantization
| Date | Title | Paper | Code | Recom |
|---|---|---|---|---|
| 2024.08 | 🔥[Transfusion] Transfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Model(@meta) | [pdf] | [transfusion-pytorch]  |
⭐️⭐️ |
| 2024.08 | 🔥[MixDQ] MixDQ: Memory-Efficient Few-Step Text-to-Image Diffusion Models with Metric-Decoupled Mixed Precision Quantization (@THU&Infinigence AI.) | [pdf] | [mixdq]  |
⭐️⭐️ |
| 2024.08 | 🔥[ViDiT-Q] ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation (@THU&Infinigence AI.) | [pdf] | [viditq]  |
⭐️⭐️ |
| 2024.08 | 🔥[VQ4DiT] VQ4DiT: Efficient Post-Training Vector Quantization for Diffusion Transformers(@ZJU) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.08 | 🔥[LBQ] Low-Bitwidth Floating Point Quantization for Efficient High-Quality Diffusion Models(@toronto.edu) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.08 | 🔥[EE-Diffusion] A Simple Early Exiting Framework for Accelerated Sampling in Diffusion Models(@KAIST AI) | [pdf] | [ee-diffusion]  |
⭐️⭐️ |
| 2024.08 | 🔥[TFM-PTQ] Temporal Feature Matters: A Framework for Diffusion Model Quantization(@SenseTime) | [pdf] | ⚠️ | ⭐️⭐️ |
| 2024.08 | 🔥[Diffusion-RWKV] Diffusion-RWKV: Scaling RWKV-Like Architectures for Diffusion Models(@Zhengcong Fei) | [pdf] | [Diffusion-RWKV]  |
⭐️⭐️ |
| 2024.09 | 🔥[LinFusion] LINFUSION: 1 GPU, 1 MINUTE, 16K IMAGE(@NUS) | [pdf] | [LinFusion]  |
⭐️⭐️ |
| 2024.11 | 🔥🔥[SVDQuant] SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models | [pdf] | [nunchaku]  |
⭐️⭐️ |
📙 Attention
| Date | Title | Paper | Code | Recom |
|---|---|---|---|---|
| 2024.10 | 🔥🔥[SageAttention] SAGEATTENTION: ACCURATE 8-BIT ATTENTION FOR PLUG-AND-PLAY INFERENCE ACCELERATION(@thu-ml) | [pdf] | [SageAttention]  |
⭐️⭐️ |
| 2024.11 | 🔥🔥[SageAttention-2] SageAttention2: Efficient Attention with Thorough Outlier Smoothing and Per-thread INT4 Quantization(@thu-ml) | [pdf] | [SageAttention] |
⭐️⭐️ |
| 2025.03 | 🔥🔥[SpargeAttention] SpargeAttn: Accurate Sparse Attention Accelerating Any Model Inference(@thu-ml) | [pdf] | [SpargeAttn]  |
⭐️⭐️ |
| 2025.05 | 🔥🔥[SageAttention-3] SageAttention3: Microscaling FP4 Attention for Inference and An Exploration of 8-bit Training(@thu-ml) | [pdf] | [SageAttention] |
⭐️⭐️ |
| 2025.05 | 🔥🔥[DraftAttention] DraftAttention: Fast Video Diffusion via Low-Resolution Attention Guidance(@Northeastern University) | [pdf] | [draft-attention]  |
⭐️⭐️ |
©️License
GNU General Public License v3.0
🎉Contribute
Welcome to star & submit a PR to this repo!