# datajuicer/data-juicer

Data processing for and with foundation models!  🍎 🍋 🌽 ➡️ ➡️🍸 🍹 🍷

Repository: https://github.com/datajuicer/data-juicer
Canonical: https://ross.abutalabs.com/products/data-juicer
Homepage: https://datajuicer.github.io/data-juicer/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: data-analysis, data-science, large-language-models, llm, data-visualization, llms, instruction-tuning, pre-training, multi-modal, synthetic-data, data, data-pipeline, data-processing, foundation-models
Last push: 2026-08-26T11:26:06+00:00

## Health v2 (maintenance only)
Score: 94/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 96, longevity 80
- inputs: {"age_days": 1128, "days_push": 7, "days_rel": 26, "gap_med": 19.5, "n_releases_24m": 25}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 6938, forks 407 (observed 2026-08-28T04:09:51.621862+00:00)

## What it is
Data-Juicer is a Python library and data processing system for cleaning, deduplicating, synthesizing, and analyzing data for foundation model training. It provides 200+ composable operators that scale from a laptop to thousand-node clusters.

## Use cases
- clean and deduplicate web-scale pre-training corpora for LLMs
- filter and curate instruction-tuning datasets
- prepare domain-specific RAG index data
- synthesize training data for foundation models
- process multimodal datasets for AI training
- analyze and visualize dataset quality before training
- curate agent interaction traces for training

## When to choose
- you need to clean, filter, or deduplicate large datasets for LLM pre-training or fine-tuning
- you want composable data processing operators with recipes for foundation model data
- you need data processing that scales from laptop to large clusters
- you are preparing multimodal or synthetic training data

## When to avoid
- you need a simple one-off ETL job unrelated to AI/ML data
- you require a fully managed GUI data platform rather than a code-first library
- your data volumes are small and a few pandas scripts would suffice

## Facets
- artifact type: library
- maturity: active
- function: etl, data-science, data-visualization, machine-learning, llm-training, rag, nlp
- domain: large-language-models, data-science, artificial-intelligence
- platform: python, cloud, cross-platform
- tags: data-processing, foundation-models, synthetic-data, data-cleaning, deduplication, multimodal, data-pipeline, pre-training-data, data-engineering, docker

## Member repositories
- datajuicer/data-juicer (main) score 94

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:51.621862+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 2026-08-29T17:41:14.375656+00:00, confidence not recorded.
  - readme: https://github.com/datajuicer/data-juicer (fetched 2026-08-28T04:09:51.621862+00:00, sha f27276109fe6)
  - homepage: https://datajuicer.github.io/data-juicer/ (fetched 2026-08-29T08:37:07.465960+00:00, sha fc6c7e5d5afd)
- Data as of 2026-08-30T08:39:29.467469+00:00.