SPICE: Sparse Parameter Gossip for LLM Pretrainingwith Daniele Kopyshevskiy, Edouard Oyallon, Eugene Belilovsky1 versionW_hg6sdk7pv1 · current
Unifying Local Communications and Local Updates for LLM PretrainingPietro Cagnasso, Eugene Belilovsky, Edouard Oyallon0 citations2606.11081v12026arXiv
Beyond Local Sharpness: Communication-Efficient Global Sharpness-aware Minimization for Federated LearningDebora Caldarola, Pietro Cagnasso, Barbara Caputo et al.9 citations2412.03752v22024arXiv