BBSRguy/ODEN-Indcorpus

ODEN‑Indcorpus 📚

ODEN‑Indcorpus is a 3.7‑million‑line Odia mixed text collection curated from fiction, dialogue, encyclopaedia, Q‑A and community writing derived from the ODEN initiative.
After thorough normalisation and de‑duplication it serves as a robust substrate for training Odia‑centric tokenizers, language models and embedding spaces.

Split	Lines
Train	3,373,817
Validation	187,434
Test	187,435
Total	3,748,686

The material ranges from conversational snippets to encyclopaedic passages and reflects both classical and contemporary Odia usage, ensuring vocabulary coverage across formal and colloquial registers.

Quick‑start

from datasets import load_dataset

ds = load_dataset("BBSRguy/ODEN-Indcorpus", split="train", streaming=True)
for line in ds.take(3):
    print(line["text"])

Intended uses

• Training Byte‑/SentencePiece tokenizers optimised for Odia
• Pre‑training or continued training of Odia‑focused LLMs / ALMs
• Embedding evaluation, topic modelling, text classification baselines

Citation

@misc{oden-indcorpus-2025,
  title  = {ODEN‑Indcorpus: A 3.7‑M line Odia Text Dataset},
  author = {@BBSRguy},
  year   = 2025,
  howpublished = {\url{https://huggingface.co/datasets/BBSRguy/ODEN-Indcorpus}}
}