8000 llm_benchmark_eval/data at main · Dominic789654/llm_benchmark_eval · GitHub
[go: up one dir, main page]

Skip to content

Latest commit

 

History

History

data

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

README.md

Benchmark Datasets Directory

This directory contains benchmark datasets for evaluation. Each dataset should be in its own subdirectory with JSONL format files.

Supported Datasets

  • MATH - Competition mathematics problems
  • GSM8K - Grade school math word problems
  • AIME - American Invitational Mathematics Examination
  • AIME25 - Recent AIME problems (2025)
  • AMC - American Mathematics Competitions
  • OlympiadBench - Mathematical olympiad problems
  • Minerva - STEM reasoning problems
  • GPQA - Graduate-level physics questions
  • MMLU-Pro - Enhanced MMLU with challenging problems

Dataset Format

Each JSONL file should contain one JSON object per line:

{"problem": "What is 2+2?", "answer": "4"}

Required Fields:

  • problem - The question or problem statement (string)
  • answer - The correct answer (string)

Optional Fields: id, difficulty, category, source, year

0