Sign InOpen Brain
arXivPaperNeeds Review

UE5M3 FP4 Block Scaling for Stable Language Model Pretraining

A UE5M3 block-scaling recipe trained an 8B model in FP4 without Hadamard transforms or BF16 final layers, while reporting better losses and downstream estimates than the compared recipe.

arXiv · Sep 2, 2026
Open Source Open MarkdownOpen JSON
Source Summary

The authors paired E2M1 payloads with **UE5M3 block scales**, used selective stochastic rounding, removed the randomized Hadamard transform, and applied FP4 to eligible internal linears. They pretrained an **8B model on nearly 190B tokens**.

Practical Implication

For low-precision training work, test whether wider block scales can replace auxiliary transforms and higher-precision exemptions. The simpler recipe may reduce work outside FP4 matrix multiplications.

Agent-Ready Context
The authors paired E2M1 payloads with **UE5M3 block scales**, used selective stochastic rounding, removed the randomized Hadamard transform, and applied FP4 to eligible internal linears. They pretrained an **8B model on nearly 190B tokens**.

For low-precision training work, test whether wider block scales can replace auxiliary transforms and higher-precision exemptions. The simpler recipe may reduce work outside FP4 matrix multiplications.

Results come from software-emulated training and recipe-specific inference policies. The reported **21.2% throughput increase** is a native execution ablation that jointly removed two components, not a direct end-to-end UE5M3 hardware result.
Connected Context · Feed7 Judgment

This adds numerical format and scaling design to model-selection judgment: a wider FP4 block scale may simplify low-precision pretraining by removing transforms and exemptions. The evidence remains recipe- and execution-specific, so the throughput figure should motivate native hardware validation rather than be treated as an end-to-end deployment gain.

Context Map
model#model-selection
Uncertainty
Results come from software-emulated training and recipe-specific inference policies. The reported **21.2% throughput increase** is a native execution ablation that jointly removed two components, not a direct end-to-end UE5M3 hardware result.