PulseAugur
EN
LIVE 10:32:36

New FOAM optimizer matches SOAP/Muon LLM training gains with half the state

By PulseAugur Editorial · [1 sources] ·

Researchers have introduced FOAM, a new variant of the SOAP optimizer designed for large language model pre-training. FOAM aims to achieve the performance benefits of second-order optimizers like SOAP and Muon while significantly reducing computational and memory overhead. It achieves this by using input-only preconditioning and a factored Adam-moment estimator, which halves the optimizer state and speeds up training time to be comparable with Muon-class models. AI

IMPACT FOAM's efficiency improvements could enable larger-scale LLM pre-training with reduced computational resources.

RANK_REASON The cluster contains an academic paper detailing a new optimization method for LLM pre-training. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Hugging Face Daily Papers →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

New FOAM optimizer matches SOAP/Muon LLM training gains with half the state

How we ranked this

Signal score
1 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
The cluster contains an academic paper detailing a new optimization method for LLM pre-training. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
paper, infra
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
1 days old
Coverage has settled into its steady-state source set.

Full methodology in our editorial standards.

COVERAGE [1]

  1. Hugging Face Daily Papers TIER_1 English(EN) ·

    Factored One-sided Adam-Moment for Practical and Scalable SOAP

    FOAM is a lightweight SOAP variant that uses input-only preconditioning and a factored Adam-moment estimator to match strong second-order and Muon baselines in LLM pre-training while using half the optimizer state and running at Muon-class speed.

RELATED ENTITIES

RELATED TOPICS