AI Paper Atlas

Research map / Optimization / Optimizers

One-Step Gradient Delay is Not a Barrier for Large-Scale Asynchronous Pipeline Parallel LLM Pretraining

Philip Zmushko, Egor Petrov, Nursultan Abdullaev, Khrushchev Mikhail, Samuel Horváth

ICML 2026 · Optimization › Optimizers · approach: Memory-efficient LLM optimizers · 0 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use AdamW instead of Muon under one-step delaydelayed training · FineWeb-Edu (360M)validation loss: 0.012 vs 0.046Hurt
add Error Feedback correctionAdamW async · FineWeb-Edu (360M)validation loss: 2.890 -> 2.640Improved
add Error Feedback correctionMuon async · FineWeb-Edu (360M)validation loss: 2.590 -> 2.583Improved
add Error Feedback correctionAdan async · FineWeb-Edu (360M)validation loss: 2.651 -> 2.653Hurt
add Error Feedback correctionNAdam async · FineWeb-Edu (360M)validation loss: 2.694 -> 2.703Hurt
add synchronous start before delayed updates (ablation)MARS async · FineWeb-Edu (360M)validation loss: 2.897 -> 2.820Improved
add synchronous start before delayed updates (ablation)Adan async · FineWeb-Edu (360M)validation loss: 2.651 -> 2.656Hurt
add synchronous cooldown near end of training (ablation)Muon async · FineWeb-Edu (135M)validation loss: 2.856 -> 2.853No clear effect
add Taylor-based DC-ASGD delay compensation (lambda 10^4 to 10^8) (ablation)Muon async · FineWeb-Edu (135M)validation lossNo clear effect
increase momentum mu from 0.95 to 0.99 (ablation)Muon with original PipeDream · FineWeb-Edu (135M)validation lossImproved
replace PipeDream with PipeDream-2BW constant delayasync pipeline schedule at P=16 · FineWeb-Edu (135M)validation loss: loses more than 0.03Improved
add Error Feedback with 10B MoEMuon async · FineWeb (10B MoE)validation loss: 1.911 -> 1.906Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy