MiniCPM-SALA: Hybridizing Sparse and Linear Attention for Efficient Long-Context Modeling Paper • 2602.11761 • Published 3 days ago • 5