ConMoo/ChatCFD_dataset

---
license: apache-2.0
---

# ChatCFD Dataset: Benchmarks and Resources for Automated CFD Analysis

## ✨ Overview

This repository is the test dataset for the [ChatCFD](https://github.com/ConMoo/ChatCFD) project. ChatCFD is an intelligent agent system driven by Large Language Models (LLM), designed to achieve end-to-end automation of Computational Fluid Dynamics (CFD) simulations through natural language instructions.

To support academic research and community development, we have open-sourced the core data used to evaluate ChatCFD's performance, including benchmark sets based on official OpenFOAM tutorials and complex simulation cases from academic papers.

## 📊 Dataset Structure

This dataset contains approximately 2.51 GB of resources, divided into three main parts:

### 1. OpenFOAM Case Descriptions (`OF_case_description/`)

Contains a series of **natural language descriptions** (typically in `.txt` format) based on official OpenFOAM tutorial cases. These descriptions simulate simulation requirements that users would input through spoken or written language in real-world scenarios, used to test the LLM agent's ability to extract physical scenarios and simulation parameters.

### 2. OpenFOAM Case Grids (`OF_case_grids/`)

Contains **OpenFOAM-format mesh files** corresponding to the basic cases mentioned above. These meshes have been validated and can be directly invoked by the ChatCFD system for simulation calculations after automatically configuring the cases.

### 3. Paper Reproduction Benchmark Cases (`papers_and_grids/`)

This directory showcases ChatCFD's ability to handle complex tasks, containing two typical cases (NACA0012, nozzle simulation):
- **Academic Papers**: Original papers (PDF format) from web sources or academic databases, providing detailed physical setup instructions.
- **Mesh Files**: `.msh` format meshes corresponding to the paper cases.

This part of the data is used to verify whether ChatCFD can directly understand complex CFD configurations from unstructured academic literature and drive simulation workflows.

## 🚀 Quick Start

This dataset can be used in conjunction with the [ChatCFD source code](https://github.com/ConMoo/ChatCFD). You can choose either of the following methods to download.

After downloading, please refer to the project configuration file guide to fill in the corresponding mesh and description file paths into the system parameters to start end-to-end automated simulation testing.

### Download Method 1: Using Git LFS

Clone the entire repository using Git LFS:

```bash
git lfs install
git clone https://huggingface.co/datasets/ConMoo/ChatCFD_dataset
```

### Download Method 2: Using Python + huggingface_hub

First, install the dependency package:

```bash
pip install huggingface_hub
```

Then run the following code to download the dataset to the `downloaded_data` folder in the current directory:

```python
from huggingface_hub import snapshot_download
import time

# Download dataset to "downloaded_data" folder in current directory
max_retries = 5
for attempt in range(max_retries):
try:
snapshot_download(
repo_id="ConMoo/ChatCFD_dataset",
repo_type="dataset",
local_dir="./downloaded_data",
max_workers=8, # Optionally reduce concurrency to avoid connection overload
)
break
except Exception as e:
if attempt < max_retries - 1:
wait_time = (attempt + 1) * 5
print(f"Download failed (attempt {attempt + 1}/{max_retries}): {e}")
print(f"Retrying in {wait_time} seconds...")
time.sleep(wait_time)
else:
raise

print("🎉 Download complete! Please check the ./downloaded_data directory.")
```

## 📄 Citation

If you use this dataset or the ChatCFD system in academic research, please cite our preprint:

```
@article{fan2026chatcfd,
title={ChatCFD: A Large Language Model-Driven Agent for End-to-End Computational Fluid Dynamics Automation with Structured Knowledge and Reasoning},
author={Fan, E and Hu, Kang and Wu, Zhuowen and Ge, Jiangyang and Miao, Jiawei and Zhang, Yuzhi and Sun, He and Wang, Weizong and Zhang, Tianhan},
journal={Advanced Intelligent Discovery},
pages={e202500174},
year={2026},
publisher={Wiley Online Library}
}
```

## 💬 Community & Contact

Project Homepage: [ChatCFD - GitHub](https://github.com/ConMoo/ChatCFD)

Contact Email: [e-mail], [e-mail]