Cayley-Math / components /google-math /validation.json
AvrovaDonz's picture
feat(dataset): rebuild google-math with the AlphaEvolve and FunSearch constructions
cff1a8e verified
Raw History Blame
2.36 kB
{
"status": "passed",
"component": "google-math",
"sources": 11,
"records": 480605,
"groups": 3424,
"record_types": {
"research_notebook": 23,
"geometry_rule_table": 2,
"formal_geometry_problem": 265,
"formal_lean_proof": 535,
"natural_language_proof": 12,
"formal_lean_statement": 2492,
"generated_qa_pair": 477276
},
"splits": {
"train": {
"records": 479337,
"groups": 3354,
"sources": 11,
"text_characters": 86369833
},
"dev": {
"records": 1233,
"groups": 35,
"sources": 7,
"text_characters": 234753
},
"test": {
"records": 35,
"groups": 35,
"sources": 5,
"text_characters": 428472
}
},
"source_files_hashed": 3068,
"packed": {
"train": {
"rows": 19676,
"tokens": 40295101,
"loss_bearing_labels": 39815764
},
"dev": {
"rows": 51,
"tokens": 103210,
"loss_bearing_labels": 101977
},
"test": {
"rows": 92,
"tokens": 186986,
"loss_bearing_labels": 186951
}
},
"checks": [
"all_delivered_file_hashes_and_sizes",
"pinned_sources_metadata_and_excluded_candidates",
"source_file_inventory_local_bytes",
"group_split_policy_recomputed_from_scratch",
"lossless_jsonl_parquet_exact_text_uniqueness_and_record_provenance",
"group_split_separation",
"verbatim_lean_source_equality_local_bytes",
"lean_and_generated_records_survive",
"per_file_extraction_audit",
"dedup_audit_present",
"generator_audit_seed_and_quota",
"all_packed_tokens_reencoded_source_shift_padding_masks_and_attribution",
"statistics_match_artifacts"
],
"mathematical_verification": "not_performed",
"lean_compilation": "not_performed",
"limitations": [
"No Lean compilation, proof checking or mathematical verification.",
"PDF extraction is lossy for symbols, ligatures and formula layout.",
"Generated QA pairs are a seeded sample of a school-level curriculum, not research mathematics and not the paper-complete dataset.",
"formal-conjectures upstream notes third-party CC-BY-SA derived content; no per-file license audit.",
"No near-duplicate or Phase 1 overlap assessment.",
"SentencePiece decoding normalises runs of spaces, inherited from the shared Cayley-Math tokenizer."
]
}