The model file only contains the models, but no information about assets or algos. The model inside the file is identified by its number. So if the number is wrong, a wrong model is assigned to an asset. This can happen for instance when you change the order of assets after training.

But I also see a potential bug, not in your snippet but outside it: If you set Spread etc. = 0 in your script, as in my article, make sure to do that for all assets. Otherwise you'll indeed get much worse results with multiple assets.