FOAM folds Adam optimizer states into block-wise gradient means with residual correction, cutting memory overhead by up to 90% while matching vanilla Adam convergence rates under standard non-convex assumptions.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
FOAM: Blocked State Folding for Memory-Efficient LLM Training
FOAM folds Adam optimizer states into block-wise gradient means with residual correction, cutting memory overhead by up to 90% while matching vanilla Adam convergence rates under standard non-convex assumptions.