RavinduPabasara/tripitaka-en-chunks
Tripitaka English Chunks 28,554 paragraph-level chunks of the Pali Canon (Tipiṭaka/Tripitaka) in English — Bhante Sujato's translations from SuttaCentral, segmented and structured for retrieval-augmented generation (RAG), semantic search, and study tools. Each chunk keeps its canonical segment ID and a deep link back to SuttaCentral, so applications can always cite the exact passage. Coverage 4,166 distinct suttas across 13 collections: Collection Chunks… See the full description on the dataset page: https://huggingface.co/datasets/RavinduPabasara/tripitaka-en-chunks.
Tripitaka English Chunks
28,554 paragraph-level chunks of the Pali Canon (Tipiṭaka/Tripitaka) in English — Bhante Sujato's translations from SuttaCentral, segmented and structured for retrieval-augmented generation (RAG), semantic search, and study tools.
Each chunk keeps its canonical segment ID and a deep link back to SuttaCentral, so applications can always cite the exact passage.
Coverage
4,166 distinct suttas across 13 collections:
The four principal Nikāyas are covered in full; Khuddaka Nikāya coverage is partial (e.g. the Dhammapada and Jātaka include only a subset).
Fields
Usage
from datasets import load_dataset
ds = load_dataset("RavinduPabasara/tripitaka-en-chunks", split="train")
dn = ds.filter(lambda r: r["collection"] == "dn")Chunks are paragraph-level (median ~180 characters); very short rows are section headings and verse lines kept to preserve the text flow — filter on len(text) if your application doesn't want them. Rows are ordered by collection, sutta, and segment.
Processing
Scraped from SuttaCentral (Bhante Sujato's English translations only), segmented at paragraph level, then: Unicode NFC-normalized, whitespace-collapsed, empty rows dropped, and de-duplicated on source_id (29,215 raw → 28,554 rows).
License and credit
Bhante Sujato's SuttaCentral translations are dedicated to the public domain (CC0 1.0); this derived dataset carries the same dedication. Please credit SuttaCentral when you use it, and consider linking back to the url field when displaying passages — the canon deserves its context.
Dataset prepared by Ravindu Pabasara Karunarathna.
