MonumentalSystems/quality-corpus-180m-v1

MonumentalSystems Quality Corpus 180M v1

The canonical default config for this dataset is the original full parquet corpus.

Additional enriched configs are included for both the full locally scored parquet corpus and a smaller JSONL subset.

Configs

default

• Full original parquet corpus
• Files are served from quality corpus/train-*.parquet and quality corpus/validation-*.parquet

enriched_full

• Full locally enriched/scored parquet corpus
• Files: 13 parquet shards
• Rows: 8,417,395
• Files are served from quality corpus_enriched/train-*.parquet and quality corpus_enriched/validation-*.parquet
• Added fields include: source_id, source_ref, source_type, split_source, domain, doremi_domain, doremi_weight, content_type, subject, reading_level, complexity, curriculum_stage, difficulty, flesch, mtld, quality_pass, quality_reason, word_count, char_count, token_count_est, checksum, normalized_checksum, alnum_checksum, dedup_key, row_id

enriched_subset

• Locally enriched/scored JSONL subset
• Files: 9 JSONL shards
• Rows: 8,679
• Added fields include: content_type, subject, doremi_domain, reading_level, complexity, curriculum_stage, difficulty, flesch, mtld, quality_pass, quality_reason, char_count, token_count_est

Notes

• The enriched_full config was produced from the local folder C:\Users\lisam\OneDrive\Documents\quality corpus_enriched.
• The enriched subset was produced from local cleaned shards under .data/master_corpus/cleaned/ms_quality_corpus_180m_enriched.
• subject and content_type in the enriched subset are heuristic inference fields and may still need rule refinement.
• subject and content_type in enriched_full are also heuristic inference fields and may still need rule refinement.