Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
A large-scale benchmark for machine learning methods in fluid dynamics
| Date | Stars |
|---|---|
| 2026-07-31 | 290 |
| 2026-08-02 | 290 |
| 2026-08-06 | 290 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
<h1 align="center">CFDBench</h1>
<div align="center">
<a href="https://huggingface.co/datasets/chen-yingfa/CFDBench">🤗Dataset</a> | <a href="https://www.preprints.org/manuscript/202309.1550/v1">Paper (preprints.org)</a> |
<a href="https://arxiv.org/abs/2310.05963">Paper (arXiv)</a>
</div>
<div align="center">
Yining Luo, Yingfa Chen, and Zhen Zhang</br>
Tsinghua University</br>
[email protected]
</div>
<div></br></div>
This is the code for the paper: [CFDBench: A Large-Scale Benchmark for Machine Learning Methods in Fluid Dynamics](https://arxiv.org/abs/2310.05963).
CFDBench is the first large-scale benchmark for evaluating machine learning methods in fluid dynamics with varied boundary conditions (BCs), physical properties, and domain geometries. It consists of four classic problems in computational fluid dynamics (CFD), with many varying operating parameters, making it perfect for testing the inference-time generalization ability of surrogate models. Such generalizability is essential for avoiding expensive re-training when applying surrogate models to new problems.
## Quick Links
- [Data](#data)
- [Models](#models)
- [Main Results](#main-results)
- [How to Run?](#how-to-run)
- [How to Add New Models/Datasets?](#how-to-add-new-modelsdatasets)
- [Citation](#citation)
## Data
Main download links:
Interpolated data (~13.4GB):
- [🤗HuggingFace](https://huggingface.co/datasets/chen-yingfa/CFDBench)
Raw data (~460GB):
- [🤗HuggingFace](https://huggingface.co/datasets/chen-yingfa/CFDBench-raw)
- [Baidu Drive (百度网盘)](https://pan.baidu.com/s/1p0q60cv2hFZ7UcIf3XKSaw?pwd=cfd4) (提取码: cfd4)
> You likely do not need the raw data if you are just trying to evaluate and/or train neural networks on our data.
### Data Generation
The directory `generation-code` contains the code for creating the mesh (ICEM code) and the schema code for batch generation in ANSYS Fluent.
> This part takes a lot of time, and you are better off using our generated data instead.
### Data Interpolation
After generating data with numerical algorithms, it is then interpolated to a grid of 64x64. The raw data before interpolation is very large; the link below is the interpolated data.
Main download link: [[click here]](https://cloud.tsinghua.edu.cn/d/435413b55dea434297d1/)
Contains 4 problems:
- `cavity`: Lid-driven cavity flow
- `tube`: Flow through a circular tube
- `dam`: Flow over a dam
- `cylinder`: Flow around a cylinder
> The cylinder flow is separated into three files because the file size exceeds the upload limit.
Each dataset includes 3 subsets, corresponding to changing BCs, domain geometries, and physical properties.
The directory tree for the datasets:
```
▼ cavity/
▼ bc/
▼ case0000/
▼ u.npy
▼ v.npy
► case0001/
► geo/
► prop/
► tube/
► dam/
► cylinder/
```
The actual data for each velocity field is stored in `u.npy` and `v.npy`.
## Models
The basic types of models are autoregressive and non-autoregressive:
- Autoregressive:
- Auto-FFN
- Auto-DeepONet
- Auto-EDeepONet
- Auto-DeepONetCNN
- ResNet
- U-Net
- FNO
- Non-autoregressive
- FFN
- DeepONet
The implementation of the models is located in `src/models`
## Main Results
### Multi-Step Inference
### Autoregressive Models
### Nonautoregressive Models
## How to Run?
Tested on:
- PyTorch 1.13.3+cu117
- Python 3.9.0
### Installation
Make sure you have access to CUDA GPU, then setup the environment using
```shell
pip install -r requirements
```
### Prepare Data
Move the downloaded data into a `data` directory next to `src` directory, such that the directory
looks like:
```
▼ data/
▼ cavity/
▼ bc/Excerpt of 7,679 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:753ad20b1d5ea4c5, topic:benchmark