README.md

metadata

task_categories:
  - text-generation
language:
  - sa
size_categories:
  - 100K<n<1M

📊 Dataset Creation & Processing

Source Datasets

We processed 9 different Sanskrit datasets containing various types of Sanskrit texts:

Dataset	Source	Approx Texts Found	Column Used
snskrt/Sanskrit_shlokas	General Sanskrit verses	194,332	text
snskrt/Vishnu_Samhita	Vishnu Samhita	2,589	Sanskrit
snskrt/Vyakaran	Tokenized Sanskrit	615,671	Sanskrit
snskrt/Ramayan	Ramayana	23,397	shlok
snskrt/Sanskrit_OCR_Parallel_Corpus	Image labels	6,265	shlok
snskrt/Mahabharat	Mahabharata	73,791	Sanskrit
snskrt/Shrimad_Bhagvat_Puran	Bhagavata Purana	14,213	Sanskrit
snskrt/Agni_Puran	Agni Purana	10,940	Sanskrit
snskrt/Bhavishya_Puran	Bhavishya Purana	26,212	Sanskrit

Data Cleaning Process

What was removed:

• ✅ English numbers (0-9)
• ✅ English letters (a-z, A-Z)
• ✅ Non-Sanskrit characters
• ✅ Duplicate texts (303,306 duplicates removed)

What was preserved:

• ✅ Sanskrit Devanagari characters (U+0900-U+097F)
• ✅ Sanskrit punctuation: । ॥ (danda, double danda)
• ✅ Common punctuation: , . : ; ! ? ( ) [ ] { } " ' - —
• ✅ Whitespace and line breaks
• ✅ Zero-width joiner/non-joiner (U+200C, U+200D)