# ERTACache

**Repository Path**: ByteDance/ERTACache

## Basic Information

- **Project Name**: ERTACache
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Apache-2.0
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-09-05
- **Last Updated**: 2025-09-08

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# ERTACache: Error Rectification and Timesteps Adjustment for Efficient Diffusion


## 🫖 Introduction 

In this work, we present **ERTACache**, a principled and efficient caching framework for accelerating diffusion model inference. By decomposing cache-induced degradation into feature shift and step amplification errors, we develop a dual-path correction strategy that combines offline-calibrated reuse scheduling, trajectory-aware timestep adjustment, and closed-form residual rectification. The following figure gives an overview of our **ERTACache** framework, which adopts a dual-dimensional correction strategy: (1) we first perform offline policy calibration by searching for a globally effective cache schedule using residual error profiling; (2) we then introduce a trajectory-aware timestep adjustment mechanism to mitigate integration drift caused by reused features; (3) finally, we propose an explicit error rectification that analytically approximates and rectifies the additive error introduced by cached outputs, enabling accurate reconstruction with negligible overhead.

![visualization](./visualize/framework.png)

As the figure shown below, **ERTACache** preserves fine-grained visual details and frame-to-frame consistency, outperforming TeaCache and matching the performance of the non-cache reference. In video generation tasks using CogVideoX, Wan2.1-1.3B, and OperaSora 1.2, ERTA-Cache achieves noticeably better temporal consistency, particularly between the first and last frames. When applied to the Flux-dev 1.0 image model, it enhances visual richness and details. These results highlight **ERTACache** as a uniquely effective solution that balances visual quality and computational efficiency for consistent video generation.

![visualization](./visualize/exp.png)

Unlike prior heuristics-based methods, **ERTACache** provides a theoretically grounded yet lightweight solution that significantly reduces redundant computations while maintaining high-fidelity outputs. Empirical results across multiple benchmarks validate its effectiveness and generality, highlighting its potential as a practical solution for efficient generative sampling.

## 🎉 Supported Models

**Text to Video**
- **ERTACache4Wan2.1**
- **ERTACache4CogVideoX-2B**
- **ERTACache4OpenSora1.2**

**Text to Image**
- **ERTACache4FLUX**






## 📈 Inference Comparisons on a Single A800
<table>
  <tr>
    <th>T2V Model</th>
    <th>Method</th>
    <th>LPIPS</th>
    <th>SSIM</th>
    <th>PSNR</th>
    <th>Latency(s)</th>
  </tr>
  <tr>
    <td rowspan="2">OpenSora 1.2</td>
    <td>TeaCache</td>
    <td>0.2511</td>
    <td>0.7477</td>
    <td>19.10</td>
    <td>19.84</td>
  </tr>
  <tr>
    <th> ERTACache </th>
    <td>0.1659</td>
    <td>0.8170</td>
    <td>22.34</td>
    <td>18.04</td>
  </tr>
  <tr>
    <td rowspan="2">CogvideoX-2B</td>
    <td>TeaCache</td>
    <td>0.2057</td>
    <td>0.7614</td>
    <td>20.97</td>
    <td>26.88</td>
  </tr>
  <tr>
    <th> ERTACache </th>
    <td>0.1012</td>
    <td>0.8702</td>
    <td>26.44</td>
    <td>26.78</td>
  </tr>
  <tr>
   <td rowspan="2">Wan2.1-1.3B</td>
    <td>TeaCache</td>
    <td>0.2913</td>
    <td>0.5685</td>
    <td>16.17</td>
    <td>99.5</td>
  </tr>
  <tr>
    <th> ERTACache </th>
    <td>0.1095</td>
    <td>0.8200</td>
    <td>23.77</td>
    <td>91.7</td>
  </tr>
  <tr>
  <td rowspan="2">FLUX-dev 1.0</td>
    <td>TeaCache</td>
    <td>0.4427</td>
    <td>0.7445</td>
    <td>16.47</td>
    <td>14.21</td>
  </tr>
  <tr>
    <th> ERTACache </th>
    <td>0.3029</td>
    <td>0.8962</td>
    <td>20.51</td>
    <td>14.01</td>
  </tr>
  <tr>
  </tr>
</table>

<br>




## Installation

The running environment set-up depends on the specific model. For example, for FLUX, you need to install the FLUX packages:

```shell
pip install --upgrade diffusers[torch] transformers protobuf tokenizers sentencepiece
```

## Usage
For all the supported models, you can enter in the specific folder (for example: go to `\ERTACache4FLUX` ), then use the following command to get the outputs saved in the `.\sample` folder

```bash
sh run.sh
```



## 💐 Acknowledgement 

This repository is built based on [VideoSys](https://github.com/NUS-HPC-AI-Lab/VideoSys), [Diffusers](https://github.com/huggingface/diffusers), [Open-Sora](https://github.com/hpcaitech/Open-Sora),  [CogVideoX](https://github.com/THUDM/CogVideo), [FLUX](https://github.com/black-forest-labs/flux), [Wan2.1](https://github.com/Wan-Video/Wan2.1), Thanks for their contributions!

## 🔒 License 

* The majority of this project is released under the MIT license as found in the [LICENSE](./LICENSE) file.
* For [VideoSys](https://github.com/NUS-HPC-AI-Lab/VideoSys), [Diffusers](https://github.com/huggingface/diffusers), [Open-Sora](https://github.com/hpcaitech/Open-Sora),  [CogVideoX](https://github.com/THUDM/CogVideo), [FLUX](https://github.com/black-forest-labs/flux), [Wan2.1](https://github.com/Wan-Video/Wan2.1), please follow their LICENSE.