Softpick: No Attention Sink, No Massive Activations with Rectified Softmax
Paper • 2504.20966 • Published • 31
Softpick
Pretrained models from the paper "Softpick: No Attention Sink, No Massive Activations with Rectified Softmax"
Pretrained models from the paper "Softpick: No Attention Sink, No Massive Activations with Rectified Softmax"
