Dataset Viewer

The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

Tantraloka Processing Pipeline

Scripts for building a multi-layered dataset from Mark Dyczkowski's 11-volume Tantraloka translation/commentary.

Pipeline

Step Script Description
1 download_pdfs.py Download 11 volumes from HF
2 ocr_brev.py Chandra OCR-2 on single GPU with MLflow
3 ocr_brev_parallel.py Distribute OCR across 8xH100
4 parse_layers.py Parse HTML to layered segments + NER
5 build_hierarchical.py Build hierarchical dataset + audio cross-ref
6 upload_dataset.py Push to HuggingFace

Layers per verse

  • devanagari — original Sanskrit verse
  • iast — IAST transliteration
  • main_meaning — primary translation (bold text)
  • commentary — extended scholarly discussion
  • footnotes — academic references
  • audio_commentary — Dyczkowski lecture transcriptions (ch1-3, 119 lectures)
  • entities — NER: text_refs, persons, deities, concepts, tattvas, iast_terms

NER entity types

Type Pattern Examples
text_refs Regex on abbreviations + verse numbers MVV 1/15-17ab, IP 2/1/5
persons Dictionary of ~30 scholars/commentators Jayaratha, Abhinavagupta, Sanderson
deities Dictionary of ~25 deity names Siva, Para, Bhairava, Kali
concepts Dictionary of ~50 philosophical terms anuttara, vimarsa, svatantrya
tattvas Regex *tattva pattern ragatattva, mayatattva
iast_terms All terms from italic tags with IAST diacritics dharana, pramana, samavesa

Infrastructure

  • OCR: Chandra OCR-2 (datalab-to/chandra-ocr-2, 5.3B) on Brev 8xH100
  • ASR: Whisper large-v3 + Bedrock Claude Haiku (143 lectures)
  • Tracking: MLflow

Related datasets

Downloads last month
84