1 paper · 1 filter
Junchen Liu, Yi Sheng
As a widely adopted model compression technique, model pruning has demonstrated strong effectiveness across various architectures. However, we observe that when sparsity exceeds a…