Model components / Dataset Processing Tools

distilabel

Framework for building synthetic-data and AI-feedback pipelines: LLM-driven generation, labeling, and preference-data construction with pluggable model backends.

Argilla/Hugging Face distilabel; latest release v1.5.3 (2025-01-28), repo active into 2026, Apache-2.0 confirmed live 2026-07-21. Used to build the Zephyr/Notus preference data and the FinePersonas dataset.

Openness

5 high confidence

5.0

license: Apache-2.0(OSI)
source: public(argilla-io/distilabel)
governance: Argilla/Hugging Face
core-gated: ungated

Fully OSI-licensed (Apache-2.0), full source public, no proprietary core.

https://github.com/argilla-io/distilabel/blob/main/LICENSE recorded 2026-07-21
Apache License Version 2.0 text

Adoption

2 high confidence

2.0

~6.5k PyPI downloads/month; established in the HF synthetic-data workflow but modest raw volume.

https://pypistats.org/packages/distilabel recorded 2026-07-21
Downloads last month: 6,510

Capability

4 medium confidence

4.0

Indirect proxy: tool quality inferred from models where distilabel-curated data was the only changed variable. Confidence capped because there is no head-to-head against other synthetic-data frameworks.

https://huggingface.co/argilla/notus-7b-v1 recorded 2026-07-21
Notus-7B (distilabel-recurated UltraFeedback): 91.42 AlpacaEval vs Zephyr 90.60
https://huggingface.co/argilla/distilabeled-Hermes-2.5-Mistral-7B recorded 2026-07-21
distilabel Orca-DPO curation (46% of data): 54.04 Nous avg vs 53.51 full-data DPO

Unchanged since 2026-07-30 (last edited, not re-checked)