About accuracy on MOSI dataset in paper

[wangzx-tech]

Can someone tell me whether the parameter settings of these examples are very different from those in the paper? I tried to verify the performance on the MOSI dataset, but except for GRU, the accuracy rate differed by nearly 10 percent. Or can anyone tell me the correct setting method? Thank you very much！