🤖 AI Summary
This work addresses the proliferation of numeric formats in machine learning hardware, which introduces elusive silent discrepancies during cross-platform model migration, necessitating a vendor-neutral, bit-level precise reference standard. The authors construct a comprehensive catalog encompassing 84 numeric formats and introduce six bit-level consistency suites. They propose a novel verification anchor based on the φ=3.0 identity, combined with SHA-256 fingerprints and unified row patterns, to achieve semantic alignment across platforms for formats such as FP8, BF16, and MXFP4, while explicitly documenting permissible deviations from the ml_dtypes specification. All resources are packaged in JSON and integrated with IEEE P3109 v3.2.0 standard mappings, and have been open-sourced on GitHub to support reproducibility and auditability of cross-platform numeric behavior.
📝 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.