DivMerge learns task-arithmetic merging weights by minimizing Jensen-Shannon divergence between each specialist model and the merged model, improving multi-task performance and scalability.
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
-
DivMerge: A divergence-based model merging method for multi-tasking
DivMerge learns task-arithmetic merging weights by minimizing Jensen-Shannon divergence between each specialist model and the merged model, improving multi-task performance and scalability.