Dataset Viewer
Auto-converted to Parquet Duplicate
The dataset viewer is not available for this split.
Parquet error: Scan size limit exceeded: attempted to read 473829362 bytes, limit is 300000000 bytes Make sure that 1. the Parquet files contain a page index to enable random access without loading entire row groups2. otherwise use smaller row-group sizes when serializing the Parquet files
Error code:   TooBigContentError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

AgriSLM Final Pre-Training Corpus

Cleaned, deduplicated, and chunked corpus for 300M parameter agriculture SLM.

Stats

  • Total chunks (training docs): 2,242,265
  • Estimated tokens: ~5.83B
  • Duplicates removed: 99,124
  • Chunk size: 2000 words (~2,600 tokens per chunk)

Sources

  • agri-slm-corpus (PubMed, EuropePMC, Wikipedia, KCC, FAO, IRRI, CIMMYT, ICRISAT) — batches 0000-0002
  • agri-slm-corpus-2 (news, state portals, extension, universities) — batches 0000-0002
  • agri-slm-corpus-3 (KrishiKosh theses x4, KCC Q&A, ICAR ePubs, APEDA, TNAU) — batches 0003+
  • agri-slm-govt-schemes (NABARD, commodity boards, state depts) — batches 0003+

Cleaning

  • Agriculture keyword filter (60+ terms)
  • Min 80 words per chunk (15 for KCC Q&A)
  • 70% ASCII, >55% alphabetic ratio

  • Boilerplate detection (repeated line filter)
  • MD5 deduplication on first 500 words of each chunk
Downloads last month
79