Datasetpublic

microsoft/XL-DocBench

XL-DocBench Evidence-grounded reasoning across hundreds or thousands of pages. Fully verified by 194 human experts. Hongchen Wei1,†,‡, Yuanzhe Wang2,†,‡, Bei Liu2,*, Yifan Yang2, Qi Dai2, Ruichun Ma2, Kai Qiu2, Yunsheng Li2, Dongdong Chen2, Chong Luo2, Zhenzhong Chen1, Baining Guo2 1Wuhan University   2Microsoft   †Equal contribution   ‡Work done during an internship at MSRA   *Project leader Project Page · Paper · Live Leaderboard… See the full description on the dataset page: https://huggingface.co/datasets/microsoft/XL-DocBench.

sourceHugging Faceotherupdated 23d agoView on Hugging Face

7likes976downloads

rootmanifest.json

manifest.json80 linesDownload Raw Back to root

1{2  "annotation_not_answerable_non_none_count": 1,3  "answer_format_counts": {4    "Float": 110,5    "Int": 209,6    "None": 188,7    "Str": 8388  },9  "cross_doc_context_document_count_distribution": {10    "10": 19,11    "12": 4,12    "2": 4,13    "3": 14,14    "4": 17,15    "5": 8,16    "6": 20,17    "7": 19,18    "8": 47,19    "9": 220  },21  "cross_doc_evidence_document_count_distribution": {22    "0": 18,23    "2": 79,24    "3": 45,25    "4": 1226  },27  "cross_doc_qa_count": 154,28  "document_count": 292,29  "domain_counts": {30    "finance_business": 329,31    "legal_regulation": 275,32    "medical_clinical": 233,33    "narrative_literature": 81,34    "scientific_academic": 189,35    "technical_engineering": 23836  },37  "empty_evidence_qa_count": 189,38  "evidence_item_count": 3044,39  "page_indexing": "Document evidence_pages use one-based PDF/release page indices. Evidence-item pages preserve annotator-supplied page references and use page_numbering=annotator_supplied; printed page labels can differ from PDF indices. Items marked annotator_rationale are review explanations rather than verbatim document excerpts. Row-level items marked unassigned_annotation are retained provenance that is not counted as released evidence. Rows released without supporting evidence have empty evidence arrays and n_evidence_documents=0.",40  "qa_count": 1345,41  "reasoning_type_counts": {42    "aggregation": 44,43    "comparison": 200,44    "compliance": 59,45    "consistency": 19,46    "counterfactual": 56,47    "coverage": 163,48    "ranking": 156,49    "reconciliation": 107,50    "reference_chain": 206,51    "set_difference": 101,52    "temporal": 50,53    "unanswerable": 18454  },55  "release_files": {56    "cross_doc_questions": "data/qa_cross_doc.jsonl",57    "documents": "data/documents.jsonl",58    "per_question_scores": "results/scores.jsonl",59    "score_summary": "results/summary.json",60    "single_doc_questions": "data/qa_single_doc.jsonl"61  },62  "release_version": "xldocbench_strict_1345_v1",63  "score_evaluator_sha256": "efc56cb25ba136e09ceaccb5d8006ea9461f71a68998f2987073387682a5e547",64  "scored_system_count": 13,65  "single_doc_qa_count": 1191,66  "unanswerable_qa_count": 188,67  "unassigned_evidence_item_count": 11,68  "url_mapping_coverage": "292/292",69  "variant_info": {70    "variant": "strict_excluding_rag_not_approved",71    "source_git_revision": "dd2aa7fcc241c6fe0b2e5b315c220460e0f2ab94",72    "match_method": "exact_source_url",73    "filter_rule": "Remove documents with RAG / Not Approved status and every QA row that references one of those documents, then prune documents left with no QA references.",74    "rag_restricted_document_count": 37,75    "orphaned_document_count": 2,76    "removed_document_count": 39,77    "removed_qa_count": 17478  }79}80