deep-symbolic-mathematics/llm-srbench

[ICML2025 Oral] LLM-SRBench: A New Benchmark for Scientific Equation Discovery with Large Language Models

32
/ 100
Emerging

This project offers a standardized way for researchers and scientists to assess how well large language models can discover scientific equations. It provides a collection of 239 problems from various scientific fields, designed to test an LLM's ability to reason and find underlying mathematical relationships in data, rather than just recall memorized formulas. The input is scientific problem descriptions and associated data, and the output is an evaluation of the LLM's performance in uncovering the correct equations. Scientists, physicists, chemists, and biologists who use LLMs for modeling and discovery would use this.

No commits in the last 6 months.

Use this if you are a researcher or scientist evaluating how effectively different large language models can perform symbolic regression to uncover fundamental scientific equations from observational data.

Not ideal if you are looking for an off-the-shelf tool to directly discover equations for your own dataset without intending to benchmark LLM capabilities.

scientific-discovery equation-modeling physics-research chemistry-research biological-modeling
No License Stale 6m No Package No Dependents
Maintenance 2 / 25
Adoption 9 / 25
Maturity 8 / 25
Community 13 / 25

How are scores calculated?

Stars

94

Forks

11

Language

Python

License

Last pushed

Jul 31, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/transformers/deep-symbolic-mathematics/llm-srbench"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.