multitask-finetuned model failed in distillation

[zhubro321]

Let me first briefly describe the background of my problem. I want to complete the training process of pretraining-finetuning-distillation. The pretraining adopted a multi-task mode, including: FeVCH and Fe-Cu. The finetuning was also multi-task finetuning, where the finetuning dataset includes the Fe-Cu in the pre-training and another Fe-Cu dataset (in order to complete the process first, the two Fe-Cu datasets are actually the same). And then, the pretraining and finetuning processes were successfully completed, so I want to distill this multitask finetuning model.
However, following the tutorial on Bohrium does not work smoothly, the content of this tutorial does not apply to distillation of multitask finetuning model. Here is my error message :

Here is my input files :
train.json
input.json
Thank you !