An 84-Format Numeric Catalog with Bit-Exact Conformance Vectors: A Vendor-Neutral Reference for FP8, BF16, MXFP4, and Microscaling Formats
A new paper introduces a comprehensive catalog of 84 numeric formats used in machine learning hardware, addressing the challenge of silent divergences when porting models across different accelerators. The catalog includes bit-exact conformance packs for various formats like FP8, BF16, and MXFP4, serving as a vendor-neutral reference. This work aims to provide a shared standard for engineers to diagnose and resolve discrepancies, ensuring greater consistency in model performance across diverse hardware. AI
IMPACT Standardizes numeric formats, potentially reducing model porting issues and improving cross-hardware compatibility for AI workloads.