📊 GraphOmni Dataset

GraphOmni provides a unified pipeline to evaluate large language models (LLMs) on graph problems. It supports a wide range of models, graph representations, prompt types, and task formulations. The dataset enables automated evaluation and result tracking across different prompting and modeling setups.

🧠 Key Features

• Supports six graph algorithm tasks:
  Connectivity, Bfsorder, Triangle, Diameter, Cycle, Shortest path
• Includes over 240,000 examples split across these six tasks.
• Designed to benchmark LLM reasoning ability on graph tasks involving structured data (graphs).
• Each task is provided as a separate dataset split.

📁 Data Format

Each entry in the dataset is a Python dictionary with the following fields:

💡 Usage Example

You can load and convert any task split into a list format using the Hugging Face datasets library:

from datasets import load_dataset

# Load a specific task split (e.g., 'connectivity')
dataset = load_dataset("G-A-I/GraphOmni", split="connectivity")

# Convert the first 3 examples to a list of dictionaries
examples = dataset.select(range(3)).to_list()
for ex in examples:
    print(ex)