DiSF selects LLM pre-training files that are maximally decorrelated in a pretrained text embedding space, improving downstream accuracy while using only 1.5% of SlimPajama files.
Diverse client selection for federated learning via submodular maximization
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
-
Combatting Dimensional Collapse in LLM Pre-Training Data via Diversified File Selection
DiSF selects LLM pre-training files that are maximally decorrelated in a pretrained text embedding space, improving downstream accuracy while using only 1.5% of SlimPajama files.