MERIT, a max-norm and element-wise trust-ratio optimizer, improves large-batch GPT-2 and Llama training and matches small-batch downstream scores at 6k batch size.
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
CONDITIONAL 1representative citing papers
citing papers explorer
-
MERIT: Maximum-normalized Element-wise Ratio for Language Model Large-batch Training
MERIT, a max-norm and element-wise trust-ratio optimizer, improves large-batch GPT-2 and Llama training and matches small-batch downstream scores at 6k batch size.