Ask for mlp in gsm8k and math

[BiorelaxA]

In reproducing your work, because processing the CoT token in gsm8k and math500 takes a long time, I wonder if you could provide the MLP files trained on these two datasets for subsequent testing.