aiqbench analysis
This paper addresses the critical need for vendor-neutral reference material by cataloging 84 numeric formats and providing bit-exact conformance packs to help engineers diagnose silent divergences during model porting across accelerators. By explicitly documenting interpretation gaps against standards like IEEE P3109 and validating against ml_dtypes, the work offers a shared ruler for verifying floating-point behavior without making accuracy claims or asserting superiority. The release of these self-contained JSON artifacts under an open license provides a practical resource for building robust, interoperable ML hardware stacks.
An 84-Format Numeric Catalog with Bit-Exact Conformance Vectors: A Vendor-Neutral Reference for FP8, BF16, MXFP4, and Microscaling Formats
arXiv:2606.09686v1 Announce Type: cross Abstract: Numeric format proliferation in machine learning hardware -- FP8 (E4M3 and E5M2), BF16, MXFP4, microscaling block formats, and dozens of research variants -- has outpaced the availability of vendor-neutral, bit-exact reference material. Engineers porting models across accelerators encounter silent divergences that are difficult to diagnose without a shared ruler. This paper describes a catalog of 84 numeric formats spanning 13 families, a suite of six bit-exact conformance packs covering GF16, MXFP4 element, BF16, FP8 E4M3, FP8 E5M2, and E8M0 block scale, and an IEEE P3109 v3.2.0 cross-walk that maps each pack to its corresponding standards-track configured format. Each pack is a self-contained JSON document with a SHA-256 fingerprint, a shared row schema, and an anchor vector that encodes 3.0 -- the identity phi^2 + 1/phi^2 = 3 -- as a cross-pack sanity check. Packs are cross-validated against ml_dtypes 0.5.4 (Google/JAX); any divergence is documented explicitly and interpreted as a spec-permitted interpretation gap rather than hidden. The work is framed as registry filling: it does not propose new formats, make model-accuracy claims, or assert superiority over any vendor's implementation. All artifacts are publicly available at https://github.com/gHashTag/t27 under an open license.
Abstract:Numeric format proliferation in machine learning hardware -- FP8 (E4M3 and E5M2), BF16, MXFP4, microscaling block formats, and dozens of research variants -- has outpaced the availability of vendor-neutral, bit-exact reference material. Engineers porting models across accelerators encounter silent divergences that are difficult to diagnose without a shared ruler. This paper describes a catalog of 84 numeric formats spanning 13 families, a suite of six bit-exact conformance packs covering GF16, MXFP4 element, BF16, FP8 E4M3, FP8 E5M2, and E8M0 block scale, and an IEEE P3109 v3.2.0 cross-walk that maps each pack to its corresponding standards-track configured format. Each pack is a self-contained JSON document with a SHA-256 fingerprint, a shared row schema, and an anchor vector that encodes 3.0 -- the identity phi^2 + 1/phi^2 = 3 -- as a cross-pack sanity check. Packs are cross-validated against ml_dtypes 0.5.4 (Google/JAX); any divergence is documented explicitly and interpreted as a spec-permitted interpretation gap rather than hidden. The work is framed as registry filling: it does not propose new formats, make model-accuracy claims, or assert superiority over any vendor's implementation. All artifacts are publicly available at this https URL under an open license.
17 pages. Source repository: this https URL tag v4.0-trinity. Paper CC BY 4.0; code MIT. ORCID 0009-0008-4294-6159
Hardware Architecture (cs.AR); Artificial Intelligence (cs.AI); Mathematical Software (cs.MS); Performance (cs.PF); Numerical Analysis (math.NA)
G.1.0; D.3.4; B.2.4
arXiv:2606.09686 [cs.AR]
(or arXiv:2606.09686v1 [cs.AR] for this version)
https://doi.org/10.48550/arXiv.2606.09686 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 316 words
Related coverage
Golden Ruler: A Numeric Format Catalog with Bit-Exact Conformance Vectors for FP8, BF16, MXFP4, and Microscaling Formats
arXiv · Sep 7, 2026

Best Embedding Models in 2026
OpenRouter · Sep 23, 2026
Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident
Hugging Face · Jul 27, 2026

Biomedical Imaging's Real Bottleneck Is the Data, Not the Model
Databricks · Oct 8, 2026