Datasets:

Modalities:
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
License:
sangraha / README.md
safikhan's picture
Update README.md
4820b18 verified
|
raw
history blame
11.1 kB
metadata
license: cc-by-4.0
task_categories:
  - text-generation
language:
  - as
  - bn
  - gu
  - en
  - hi
  - kn
  - ks
  - ml
  - mr
  - ne
  - or
  - pa
  - sa
  - sd
  - ta
  - te
  - ur
tags:
  - language-modeling
  - casual-lm
  - llm
pretty_name: sangraha
dataset_info:
  - config_name: verified
    features:
      - name: doc_id
        dtype: string
      - name: type
        dtype: string
      - name: text
        dtype: string
    splits:
      - name: asm
      - name: ben
      - name: brx
      - name: doi
      - name: eng
      - name: gom
      - name: guj
      - name: hin
      - name: kan
      - name: kas
      - name: mai
      - name: mal
      - name: mar
      - name: mni
      - name: nep
      - name: ori
      - name: pan
      - name: san
      - name: sat
      - name: snd
      - name: tam
      - name: tel
      - name: urd
  - config_name: unverified
    features:
      - name: doc_id
        dtype: string
      - name: text
        dtype: string
    splits:
      - name: asm
      - name: ben
      - name: guj
      - name: hin
      - name: kan
      - name: mal
      - name: mar
      - name: nep
      - name: ori
      - name: pan
      - name: san
      - name: tam
      - name: tel
      - name: urd
  - config_name: synthetic
    features:
      - name: doc_id
        dtype: string
      - name: text
        dtype: string
    splits:
      - name: asm_Beng
      - name: asm_Latn
      - name: ben_Beng
      - name: ben_Latn
      - name: guj_Gujr
      - name: guj_Latn
      - name: hin_Deva
      - name: hin_Latn
      - name: kan_Knda
      - name: kan_Latn
      - name: mal_Mlym
      - name: mal_Latn
      - name: mar_Deva
      - name: mar_Latn
      - name: npi_Deva
      - name: npi_Latn
      - name: ory_Orya
      - name: ory_Latn
      - name: pan_Guru
      - name: pan_Latn
      - name: san_Deva
      - name: san_Latn
      - name: tam_Taml
      - name: tam_Latn
      - name: tel_Telu
      - name: tel_Latn
      - name: urd_Arab
      - name: urd_Latn
configs:
  - config_name: verified
    data_files:
      - split: asm
        path: verified/asm/*.parquet
      - split: ben
        path: verified/ben/*.parquet
      - split: brx
        path: verified/brx/*.parquet
      - split: doi
        path: verified/doi/*.parquet
      - split: eng
        path: verified/eng/*.parquet
      - split: gom
        path: verified/gom/*.parquet
      - split: guj
        path: verified/guj/*.parquet
      - split: hin
        path: verified/hin/*.parquet
      - split: kan
        path: verified/kan/*.parquet
      - split: kas
        path: verified/kas/*.parquet
      - split: mai
        path: verified/mai/*.parquet
      - split: mal
        path: verified/mal/*.parquet
      - split: mar
        path: verified/mar/*.parquet
      - split: mni
        path: verified/mni/*.parquet
      - split: nep
        path: verified/nep/*.parquet
      - split: ori
        path: verified/ori/*.parquet
      - split: pan
        path: verified/pan/*.parquet
      - split: san
        path: verified/san/*.parquet
      - split: sat
        path: verified/sat/*.parquet
      - split: snd
        path: verified/snd/*.parquet
      - split: tam
        path: verified/tam/*.parquet
      - split: tel
        path: verified/tel/*.parquet
      - split: urd
        path: verified/urd/*.parquet
  - config_name: unverified
    data_files:
      - split: asm
        path: unverified/asm/*.parquet
      - split: ben
        path: unverified/ben/*.parquet
      - split: guj
        path: unverified/guj/*.parquet
      - split: hin
        path: unverified/hin/*.parquet
      - split: kan
        path: unverified/kan/*.parquet
      - split: mal
        path: unverified/mal/*.parquet
      - split: mar
        path: unverified/mar/*.parquet
      - split: nep
        path: unverified/nep/*.parquet
      - split: ori
        path: unverified/ori/*.parquet
      - split: pan
        path: unverified/pan/*.parquet
      - split: san
        path: unverified/san/*.parquet
      - split: tam
        path: unverified/tam/*.parquet
      - split: tel
        path: unverified/tel/*.parquet
      - split: urd
        path: unverified/urd/*.parquet
  - config_name: synthetic
    data_files:
      - split: asm_Beng
        path: synthetic/asm_Beng/*.parquet
      - split: asm_Latn
        path: synthetic/asm_Latn/*.parquet
      - split: ben_Beng
        path: synthetic/ben_Beng/*.parquet
      - split: ben_Latn
        path: synthetic/ben_Latn/*.parquet
      - split: guj_Gujr
        path: synthetic/guj_Gujr/*.parquet
      - split: guj_Latn
        path: synthetic/guj_Latn/*.parquet
      - split: hin_Deva
        path: synthetic/hin_Deva/*.parquet
      - split: hin_Latn
        path: synthetic/hin_Latn/*.parquet
      - split: kan_Knda
        path: synthetic/kan_Knda/*.parquet
      - split: kan_Latn
        path: synthetic/kan_Latn/*.parquet
      - split: mal_Mlym
        path: synthetic/mal_Mlym/*.parquet
      - split: mal_Latn
        path: synthetic/mal_Latn/*.parquet
      - split: mar_Deva
        path: synthetic/mar_Deva/*.parquet
      - split: mar_Latn
        path: synthetic/mar_Latn/*.parquet
      - split: npi_Deva
        path: synthetic/npi_Deva/*.parquet
      - split: npi_Latn
        path: synthetic/npi_Latn/*.parquet
      - split: ory_Orya
        path: synthetic/ory_Orya/*.parquet
      - split: ory_Latn
        path: synthetic/ory_Latn/*.parquet
      - split: pan_Guru
        path: synthetic/pan_Guru/*.parquet
      - split: pan_Latn
        path: synthetic/pan_Latn/*.parquet
      - split: san_Deva
        path: synthetic/san_Deva/*.parquet
      - split: san_Latn
        path: synthetic/san_Latn/*.parquet
      - split: tam_Taml
        path: synthetic/tam_Taml/*.parquet
      - split: tam_Latn
        path: synthetic/tam_Latn/*.parquet
      - split: tel_Telu
        path: synthetic/tel_Telu/*.parquet
      - split: tel_Latn
        path: synthetic/tel_Latn/*.parquet
      - split: urd_Arab
        path: synthetic/urd_Arab/*.parquet
      - split: urd_Latn
        path: synthetic/urd_Latn/*.parquet
size_categories:
  - 100B<n<1T

Sangraha

Sangraha is the largest high-quality, cleaned Indic language pretraining data containing 251B tokens summed up over 22 languages, extracted from curated sources, existing multilingual corpora and large scale translations.

Coming Soon:

  • Sangraha Synthetic - Translated and Romanised English Wikimedia data.
  • Sangraha Verified - Hindi YouTube transcribed data.

More information:

  • For detailed information on the curation and cleaning process of Sangraha, please checkout our paper on Arxiv;
  • Check out the scraping and cleaning pipelines used to curate Sangraha on GitHub;

Getting Started

For downloading the entire Sangraha:

from datasets import load_dataset

dataset = load_dataset("ai4bharat/sangraha")

For downloading a subset (Verified/Unverified) of Sangraha:

from datasets import load_dataset

dataset = load_dataset("ai4bharat/sangraha", data_dir="<subset_name>")
# for example: dataset = load_dataset("ai4bharat/sangraha", data_dir="verified")

For downloading one language from a subset of Sangraha:

from datasets import load_dataset

dataset = load_dataset("ai4bharat/sangraha", data_dir="<subset_name>/<lang_code>")
# for example: dataset = load_dataset("ai4bharat/sangraha", data_dir="verified/asm")

Background

Sangraha contains three broad components:

  • Sangraha Verified: Containing scraped data from "human-verified" Websites, OCR-extracted data from high quality Indic language PDFs, transcribed data from various Indic language videos, podcasts, movies, courses, etc.
  • Sangraha Unverfied: High quality Indic language data extracted from existing multilingual corpora employing perplexity filtering using n-gram language models trained on Sangraha Verified.
  • Sangraha Synthetic: WikiMedia English translated to 14 Indic languages and further "romanised" from 14 languages by transliteration to English.

Data Statistics

Lang Code Verified Synthetic Unverified Total Tokens (in Millions)
asm 292.1 11,696.4 17.5 12,006.0
ben 10,604.4 13,814.1 5,608.8 30,027.5
brx 1.5 - - 1.5
doi 0.06 - - 0.06
eng 12,759.9 - - 12,759.9
gom 10.1 - - 10.1
guj 3,647.9 12,934.5 597.0 17,179.4
hin 12,617.3 9,578.7 12,348.3 34,544.3
kan 1,778.3 12,087.4 388.8 14,254.5
kas 0.5 - - 0.5
mai 14.6 - - 14.6
mal 2,730.8 13,130.0 547.8 16,408.6
mar 2,827.0 10,816.7 652.1 14,295.8
mni 7.4 - - 7.4
npi 1,822.5 10,588.7 485.5 12,896.7
ori 1,177.1 11,338.0 23.7 12,538.8
pan 1,075.3 9,969.6 136.9 11,181.8
san 1,329.0 13,553.5 9.8 14,892.3
sat 0.3 - - 0.3
snd 258.2 - - 258.2
tam 3,985.1 11,859.3 1,515.9 17,360.3
urd 3,658.1 9,415.8 1,328.2 14,402.1
tel 3,706.8 11,924.5 647.4 16,278.7
Total 64,306.1 162,707.9 24,307.7 251,321.0

To cite Sangraha, please use:

@article{khan2024indicllmsuite,
  title   = {IndicLLMSuite: A Blueprint for Creating Pre-training and Fine-Tuning Datasets for Indian Languages},
  author  = {Mohammed Safi Ur Rahman Khan and Priyam Mehta and Ananth Sankar and Umashankar Kumaravelan and Sumanth Doddapaneni and Suriyaprasaad G and Varun Balan G and Sparsh Jain and Anoop Kunchukuttan and Pratyush Kumar and Raj Dabre and Mitesh M. Khapra},
  year    = {2024},
  journal = {arXiv preprint arXiv: 2403.06350}
}