Can I cycle through multiple baseline models?

I have multiple models, and I want to train one by one in the same program, but I find that this is not the same as the result of training one model at a time. I del the model variables, but still have an impact on the new model of the next training
such as
modellist = [LSTM,GRU,CNN]
for model in modellist:
train model
test model
But,In this case, the model loss will be very different from the model trained separately under the same random seed

Create a dictionary instead of list. Then you can iterate over the keys and access the models and train them