Back to Gap Map Model components / Dataset Processing Tools

RedPajama-Data

Together AI
open source / Overall score: 2.8

The RedPajama construction and quality-signal pipeline: quality classifiers, per-document quality/toxicity/importance annotations, and exact + fuzzy deduplication. Produced RedPajama-Data-v2 (30T tokens, 5 languages).

The construction pipeline, not the dataset, behind RedPajama-v2.

Openness

5 high confidence
5.0
license
Apache-2.0(OSI)
source
public
core features withheld
no

Fully OSI-licensed (Apache-2.0), full source public.

Adoption

2 medium confidence
2.0

This project has no PyPI package, so GitHub stars are the only usage signal published for it, and a star is not a use. Development has stopped, but the project is still widely referenced as the RedPajama-v2 quality-signal pipeline.

Capability

4 medium confidence
4.0

Comprehensive quality-signal pipeline that produced a frontier-scale open corpus.

  • https://raw.githubusercontent.com/togethercomputer/RedPajama-Data/main/README.md recorded 2026-09-29

    README documents quality signals and dedup: "Here we provide code to run exact and fuzzy" deduplication, "Exact Deduplication using a Bloomfilter", "Fuzzy Deduplication with Locality Sensitive Hashing", quality signal "ccnet_perplexity", "rps_doc_ml_wikiref_score" classifier, importance weights and "rps_doc_ldnoobw_words" blocklist; title "an Open Dataset with 30 Trillion Tokens" and languages "English, German, French, Italian, Spanish".

Verified 2026-09-27