# huggingface/datatrove

Freeing data processing from scripting madness by providing a set of platform-agnostic customizable pipeline processing blocks.

Repository: https://github.com/huggingface/datatrove
Canonical: https://ross.abutalabs.com/products/datatrove
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-08-13T16:04:51+00:00

## Health v2 (maintenance only)
Score: 86/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 97, release rhythm 73, longevity 84
- inputs: {"age_days": 1176, "days_push": 20, "days_rel": 20, "gap_med": 121.5, "n_releases_24m": 7}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3308, forks 292 (observed 2026-08-28T04:07:55.913681+00:00)

## What it is
DataTrove is a Python library from Hugging Face for processing, filtering, and deduplicating large-scale text data through customizable pipeline blocks. Pipelines are platform-agnostic, running locally, on Slurm, or on Ray, making it well suited for preparing LLM training data.

## Use cases
- filter and deduplicate large text datasets for LLM training
- process common crawl web data at scale
- extract text from warc/arc/wet files
- run data processing pipelines on a slurm cluster
- build custom data processing pipeline blocks
- generate synthetic training data with LLM inference

## When to choose
- you need to clean, filter, or deduplicate massive text corpora
- you want platform-agnostic pipelines that run locally or on Slurm/Ray
- you are preparing pretraining data for large language models
- you need to read web archive formats like warc, arc, or wet

## When to avoid
- you only need small-scale data wrangling that pandas handles fine
- you need a GUI or no-code ETL tool
- your data is not text-based

## Facets
- artifact type: library
- maturity: active
- function: etl, data-science, nlp, llm-training, streaming
- domain: machine-learning, large-language-models
- platform: python, cli
- tags: data-pipelines, text-processing, deduplication, slurm, web-crawl-processing, fsspec, data-engineering, natural-language-processing, linux, docker

## Member repositories
- huggingface/datatrove (main) score 86

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:55.913681+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 2026-08-29T18:42:37.640794+00:00, confidence not recorded.
  - readme: https://github.com/huggingface/datatrove (fetched 2026-08-28T04:07:55.913681+00:00, sha 6e3f95a7ea0f)
  - registry_pypi: https://pypi.org/pypi/datatrove/json (fetched 2026-08-29T09:35:13.688200+00:00, sha 2b852b80d34a)
- Data as of 2026-08-30T08:39:29.467469+00:00.