{ "@context": { "@language": "en", "@vocab": "https://schema.org/", "sc": "https://schema.org/", "cr": "http://mlcommons.org/croissant/", "rai": "http://mlcommons.org/croissant/RAI/", "data": {"@id": "cr:data", "@type": "@json"}, "dataType": {"@id": "cr:dataType", "@type": "@vocab"}, "extract": "cr:extract", "field": "cr:field", "fileObject": "cr:fileObject", "fileSet": "cr:fileSet", "format": "cr:format", "includes": "cr:includes", "isLiveDataset": "cr:isLiveDataset", "jsonPath": "cr:jsonPath", "key": "cr:key", "md5": "cr:md5", "path": "cr:path", "recordSet": "cr:recordSet", "sha256": "cr:sha256", "source": "cr:source" }, "@type": "sc:Dataset", "conformsTo": "http://mlcommons.org/croissant/1.0", "name": "numeric-conformance-packs", "description": "Bit-exact conformance test vectors for six numeric formats used in modern ML compute (gf16, mxfp4, bf16, fp8_e4m3, fp8_e5m2, e8m0_block). Each pack contains the format spec, hand-picked vectors (canonical values, edge cases, subnormals, infinities/NaN where applicable), expected bit-pattern + decoded f64 + honest absolute error. Cross-validated against ml_dtypes 0.5.4 (Google/JAX) on overlapping formats; reconciled with IEEE P3109 v3.2.0.", "url": "https://huggingface.co/datasets/playra/numeric-conformance-packs", "version": "2.1.0", "datePublished": "2026-06-13", "creator": { "@type": "Person", "name": "Dmitrii Vasilev", "identifier": "https://orcid.org/0009-0008-4294-6159", "email": "admin@t27.ai", "affiliation": { "@type": "Organization", "name": "Trinity S3AI (independent research collective)" } }, "publisher": { "@type": "Organization", "name": "Trinity S3AI" }, "license": "https://creativecommons.org/licenses/by/4.0/", "citeAs": "@misc{vasilev2026catalog, author = {Vasilev, Dmitrii}, title = {An 84-Format Numeric Catalog with Bit-Exact Conformance Vectors}, year = {2026}, eprint = {2606.09686}, archivePrefix = {arXiv}, primaryClass = {cs.MS}, doi = {10.48550/arXiv.2606.09686}}", "keywords": [ "conformance vectors", "FP8 E4M3", "FP8 E5M2", "MXFP4", "bfloat16", "GoldenFloat GF16", "OCP Microscaling", "IEEE P3109" ], "isLiveDataset": false, "distribution": [ { "@type": "cr:FileSet", "@id": "pack_files", "name": "pack_files", "description": "One JSON file per conformance pack.", "encodingFormat": "application/json", "includes": "*_conformance_v0.json" } ], "recordSet": [ { "@type": "cr:RecordSet", "@id": "vectors", "name": "vectors", "description": "One row per conformance test vector across all six packs.", "field": [ {"@type": "cr:Field", "@id": "vectors/name", "name": "name", "description": "Vector name (e.g. zero_pos, anchor_3, max_finite).", "dataType": "sc:Text"}, {"@type": "cr:Field", "@id": "vectors/input_f64", "name": "input_f64", "description": "Source f64 input value.", "dataType": "sc:Float"}, {"@type": "cr:Field", "@id": "vectors/bits_hex", "name": "bits_hex", "description": "Encoded bit-pattern, hex string.", "dataType": "sc:Text"}, {"@type": "cr:Field", "@id": "vectors/decoded_f64", "name": "decoded_f64", "description": "Round-trip decoded f64 value.", "dataType": "sc:Float"}, {"@type": "cr:Field", "@id": "vectors/abs_error", "name": "abs_error", "description": "Honest |input - decoded|; non-zero where the input is not representable.", "dataType": "sc:Float"}, {"@type": "cr:Field", "@id": "vectors/category", "name": "category", "description": "anchor, subnormal, max_finite, special (inf/nan), edge.", "dataType": "sc:Text"} ] } ] }