#WINA31/05/2025
Microsoft’s WINA: Revolutionizing Efficient Inference for Large Language Models Without Training
Microsoft and collaborators introduce WINA, a novel training-free sparse activation method that significantly improves efficiency and accuracy in large language model inference by leveraging both neuron activations and weight norms.