README.md

metadata

pretty_name: QFT-Bench
language:
  - en
license: cc-by-4.0
task_categories:
  - question-answering
size_categories:
  - n<1K
tags:
  - physics
  - quantum-field-theory
  - high-energy-physics
  - reasoning
  - benchmark
  - llm-evaluation

LLMs Can Tackle High Energy Theoretical Physics, yet Gaps Remain:
Part I, Quantum Field Theory

📄 Paper | 🤗 Data

QFT-Bench is a benchmark designed to evaluate the reasoning capabilities of large language models (LLMs) in quantum field theory (QFT), a fundamental theoretical framework underlying modern particle physics.

The benchmark consists of 75 problems covering eight QFT core areas:

• Classical Field Theory (Classical)
• Quantization of Scalar, Spinor, and Vector Fields (Quantization)
• Quantum Electrodynamics (QED)
• Regularization and Renormalization (Renormalization)
• Non-Abelian Gauge Theory and QCD (Non-Abelian)
• Electroweak Theory and the Higgs Mechanism (Electroweak)
• Supersymmetric Field Theories (SUSY)
• Conformal Field Theories (CFT)

Dataset Structure

The dataset is provided in test.json. Each problem contains the following fields:

Field	Description
num	Problem number
question	Problem statement
category	QFT topic category
reference_answer	Reference final answer
reference_solution	Reference derivation and solution
numeric_tolerance, reference_answer_alt, specific_requirement	Additional evaluation instructions when applicable

Mathematical expressions are represented using LaTeX.

Load the dataset

from datasets import load_dataset

dataset = load_dataset("LLMsHEP/QFT-Bench", split="test")

print(f"Number of problems: {len(dataset)}")
print(dataset[0]["question"])

Citation

If you find it useful, feel free to cite our paper (that's an encouragement to us):

@article{llmsqft,
  title   = {LLMs Can Tackle High Energy Theoretical Physics, yet Gaps Remain: Part I, Quantum Field Theory},
  author  = {Liu, Tong and Li, Bohan and Cheng, Gongrui and Yu, Zixiong and He, Yulan and Tresp, Volker},
  journal = {arXiv preprint arXiv:2610.33134},
  url     = {https://qft-bench.github.io/},
  year    = {2026}
}

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.