Math 2.0

Resources / preprint

Why Does Adaptive Batching Help LLM Pretraining? A Perspective from Unbounded Variance 1AXXV9ZT11

URIm2://res/1AXXV9ZT11
Kindpreprint
CreatorsArda Fazla, Antesh Upadhyay, Ege C. Kaya, M. Berk Sahin, Abolfazl Hashemi
Date2026-10-01
Subjectsarxiv cs.LG, arxiv math.OC
Scopeborder

Expressions

e1: version, preprint

Date2026-10-01
IdsarXiv 2610.02355

Items in this resource

The mentions of the inventory that name this resource. Each name links to the card of its inventory row.

ItemKindWhereSource
Convergence of SGDM with Dynamic Batchingstatementtheorem 3arxiv
SGD Upper Bound with Dynamic Batchingstatementtheorem 2arxiv
Unrestricted algorithm reduction of the activated BG-$a$ chainstatementlemma 5arxiv
Lower bound under BG-$a$ noisestatementtheorem 1arxiv
SGDM Upper Bound with Dynamic Batchingstatementtheorem 3arxiv
Masked oracle properties from Lemma 3statementlemma 3arxiv
Deterministic chain properties (Lemma 2 of restatedstatementlemma 1arxiv
Range boundstatementlemma 2arxiv
Deterministic chaindefinitiondefinition 3arxiv
Expected SFO cost for $a>2$statementproposition 1arxiv
Lemma 6 of restatedstatementlemma 4arxiv
Lower bound under smoothnessstatementtheorem 1arxiv
Zero-respecting algorithmdefinitiondefinition 2arxiv
Convergence of SGD with Dynamic Batchingstatementtheorem 2arxiv
Progress functiondefinitiondefinition 1arxiv
Bernoulli masked chain oracledefinitiondefinition 4arxiv

Catalogue records