LeoLai689/SimulCost-full-sim_res

SimulCost-Bench Simulation Results

Project Page | Paper | GitHub | Tools

This repository stores pre-cached simulation result archives (.zip/.tar.gz) for SimulCost, a cost-aware benchmark and toolkit for evaluating how well LLM agents tune simulation parameters under realistic computational budgets.

Unlike prior evaluations that focus on correctness while implicitly treating tool usage as “free,” SimulCost explicitly measures both: (1) whether a proposed configuration meets an accuracy target and (2) how much simulation compute it consumes.

Dataset Description

The benchmark covers 12 physics simulators across fluid dynamics, solid mechanics, and plasma physics, with 2,916 single-round (initial guess) and 1,900 multi-round (trial-and-error) tasks. The simulators include:

• Fluid Dynamics: 1D Burgers, 1D/2D Euler, 2D Navier-Stokes (Channel and Transient)
• Heat Transfer: 1D Heat Transfer, 2D Steady Heat Transfer
• Plasma Physics: 1D EPOCH PIC, Hasegawa-Mima (Linear and Nonlinear)
• Solid Mechanics: 2D MPM, 2D FEM, 1D Diffusion-Reaction

This specific repository provides baseline or full simulation caches to help users skip long simulation runtimes when evaluating models.

Usage

Pre-cached Results

To use these cached results with the SimulCost toolkit:

1. Download the .zip or .tar.gz files corresponding to the simulations you need.
2. Place and extract them into your simulation results directory (default is ./sim_res/ within the toolkit directory).

Sample Inference

Once the environment is set up and results are cached, you can run inference using the toolkit provided in the GitHub repository:

# Example: Running inference on Heat 1D dataset with a provider
python inference/langchain_LLM.py -p openai -m gpt-5-2025-08-07 -d heat_1d -t cfl -l medium -z

Parameters:

• -p: LLM provider (openai, gemini, bedrock, custom_model)
• -m: Model name/identifier
• -d: Dataset name
• -t: Problem task type
• -l: Precision level
• -z: Enable zero-shot mode

Citation

If you find this benchmark or the simulation results useful in your research, please cite:

@article{cao2025simulcost,
  title={SimulCost: A Cost-Aware Benchmark and Toolkit for Automating Physics Simulations with LLMs},
  author={Cao, Yadi and Lai, Sicheng and Huang, Jiahe and Zhang, Yang and Lawrence, Zach and Bhakta, Rohan and Thomas, Izzy F. and Cao, Mingyun and Tsai, Chung-Hao and Zhou, Zihao and Zhao, Yidong and Liu, Hao and Marinoni, Alessandro and Arefiev, Alexey and Yu, Rose},
  journal={arXiv preprint arXiv:2603.20253},
  year={2025}
}