Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free

MoE
Re-assign community