llmsql-bench/llmsql-benchmark-finetune-ready

LLMSQL Benchmark (Finetune-Ready)

⚠️ A newer version of this dataset is available:
👉 https://huggingface.co/datasets/llmsql-bench/llmsql-2.0-fine-tune-ready

This benchmark is designed to evaluate text-to-SQL models. For usage of this benchmark see https://github.com/LLMSQL/llmsql-benchmark.

This repository contains a finetune-ready version of the LLMSQL benchmark: LLMSQL on Hugging Face.

The dataset is structured in a messages format suitable for instruction-tuned models, where each example has a messages field. This field is a list of dictionaries with:

• "role": "user" — the input question or prompt
• "role": "assistant" — the expected SQL query

---

Files / Subsets

The dataset is organized into three shot-based subsets:

Subset	Description	Splits
0shot	Zero-shot examples	train, validation, test
1shot	One-shot examples	train, validation, test
5shot	Five-shot examples	train, validation, test

Each subset folder contains a DatasetDict saved in Hugging Face format, which can be loaded using:

from datasets import load_dataset

# Load 1-shot subset
ds = load_dataset("llmsql-bench/llmsql-benchmark-finetune-ready", name="1shot")
print(ds["train"][0])

---

Citation

If you use this benchmark, please cite:

@inproceedings{llmsql_bench,
  title={LLMSQL: Upgrading WikiSQL for the LLM Era of Text-to-SQL},
  author={Pihulski, Dzmitry and Charchut, Karol and Novogrodskaia, Viktoria and Koco{'n}, Jan},
  booktitle={2025 IEEE International Conference on Data Mining Workshops (ICDMW)},
  year={2025},
  organization={IEEE}
}