We propose CoreMatching, a co-adaptive sparse inference framework, which leverages the synergy between token and neuron sparsity to enhance inference efficiency. For the first time, we theoretically ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results