# meta-llama/synthetic-data-kit

Tool for generating high quality Synthetic datasets

Repository: https://github.com/meta-llama/synthetic-data-kit
Canonical: https://ross.abutalabs.com/products/synthetic-data-kit
Homepage: https://pypi.org/project/synthetic-data-kit/
Language: Python
License: MIT
License Family: permissive
Topics: data, generation, llm, python, synthetic
Last push: 2025-10-28T20:10:55+00:00

## Health v2 (maintenance only)
Score: 42/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 49, release rhythm 35, longevity 37
- inputs: {"age_days": 524, "days_push": 309, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1632, forks 236 (observed 2026-08-28T04:05:14.214679+00:00)

## What it is
A Python CLI tool from Meta for generating high-quality synthetic datasets to fine-tune LLMs. It follows a four-stage pipeline (ingest, create, curate, save-as) that converts documents into QA pairs, chain-of-thought reasoning traces, and summaries in fine-tuning-friendly formats.

## Use cases
- generate synthetic QA pairs from pdfs for fine-tuning
- create chain-of-thought reasoning traces with an LLM
- convert documents into fine-tuning friendly chat formats
- curate synthetic datasets using LLM-as-a-judge
- prepare training data for Llama fine-tuning
- ingest pdfs, html, youtube transcripts, docx and ppt files
- export datasets to formats required by fine-tuning packages

## When to choose
- you need structured training data in user/assistant format for LLM fine-tuning
- you want an end-to-end pipeline from raw documents to curated fine-tuning datasets
- you use vLLM or a local/external LLM endpoint and want modular dataset generation
- you want LLM-as-a-judge curation to filter low-quality examples

## When to avoid
- you need general-purpose data labeling or annotation for non-LLM ML tasks
- you want a GUI or web interface rather than a CLI workflow
- you don't have access to an LLM backend for generation
- you need real-time or streaming data generation rather than batch pipelines

## Facets
- artifact type: cli-tool
- maturity: active
- function: data-generation, llm-training, etl, cli, nlp
- domain: machine-learning, large-language-models, developer-tools
- platform: python, cli, cross-platform
- tags: synthetic-data, fine-tuning, qa-pairs, chain-of-thought, llama, vllm, data-curation, lance-format, data-engineering

## Member repositories
- meta-llama/synthetic-data-kit (main) score 42

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:14.214679+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:47:47.448408+00:00, confidence not recorded.
  - readme: https://github.com/meta-llama/synthetic-data-kit (fetched 2026-08-28T04:05:14.214679+00:00, sha ca53158cc080)
  - homepage: https://pypi.org/project/synthetic-data-kit/ (fetched 2026-08-29T11:20:33.633588+00:00, sha 4b4e8fead74a)
  - registry_pypi: https://pypi.org/pypi/synthetic-data-kit/json (fetched 2026-08-29T11:20:33.636330+00:00, sha 6b56b0fc52c2)
- Data as of 2026-08-30T08:39:29.467469+00:00.