KeyError: 'optimizer_state_dict' arise with Pretrined model

[SHAIIM04]

I applied experimental on my own data using the pre-train model "resnet18_mstcn_video" only I modified the run as you mentioned in "You can pass the checkpoint path (.pth or .pth.tar) to the variable argument --model-path, and specify the --init-epoch to 1 to resume training." also i modified "allow_size_mismatch = True" . Then when running the code we got error .

so what is the problem and can fix this problem ?

Thanks
best regrds