ziegler-ingo/CRAFT
[TACL, EMNLP 2025 Oral] Code, datasets, and checkpoints for the paper "CRAFT Your Dataset: Task-Specific Synthetic Dataset Generation Through Corpus Retrieval and Augmentation"
This project helps machine learning practitioners create high-quality synthetic datasets to train large language models for specific tasks. It takes a small set of human-written examples (few-shots) and a large text corpus, then generates a much larger, task-specific dataset. This is ideal for machine learning engineers, data scientists, or researchers who need to fine-tune LLMs but lack sufficient real-world training data.
Use this if you need to fine-tune a large language model for a particular question-answering, summarization, or recipe generation task, but don't have enough labeled training data.
Not ideal if you're looking for a completely automated, zero-shot solution for highly specialized tasks without providing any task-specific examples.
Stars
34
Forks
10
Language
Python
License
Apache-2.0
Category
Last pushed
Dec 05, 2025
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/llm-tools/ziegler-ingo/CRAFT"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
InternScience/GraphGen
GraphGen: Enhancing Supervised Fine-Tuning for LLMs with Knowledge-Driven Synthetic Data Generation
timothepearce/synda
A CLI for generating synthetic data
rasinmuhammed/misata
High-performance open-source synthetic data engine. Uses LLMs for schema design and vectorized...
ZhuLinsen/FastDatasets
A powerful tool for creating high-quality training datasets for Large Language Models...
BatsResearch/bonito
A lightweight library for generating synthetic instruction tuning datasets for your data without GPT.