meta-llama/synthetic-data-kit

Tool for generating high quality Synthetic datasets

56
/ 100
Established

Supports multi-format document ingestion (PDF, DOCX, HTML, YouTube transcripts) and generates structured fine-tuning datasets through a modular 4-stage pipeline: ingest → create (QA pairs, Chain-of-Thought reasoning, or summaries) → curate (using Llama-as-judge quality filtering) → save-as (converts to Alpaca, OpenAI, or HuggingFace formats). Uses Lance vector storage by default and integrates with vLLM or external LLM APIs for generation, with full YAML-based configuration overrides.

1,524 stars. Available on PyPI.

Maintenance 6 / 25
Adoption 10 / 25
Maturity 18 / 25
Community 22 / 25

How are scores calculated?

Stars

1,524

Forks

215

Language

Python

License

MIT

Last pushed

Oct 28, 2025

Commits (30d)

0

Dependencies

15

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/meta-llama/synthetic-data-kit"

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