Is there a downside to loading the pre-training model？

1

Is there a downside to loading the pre-training model？

When we are training the model， Load the pre-training model can get the following advantage:

  1. Accelerated training, and can use less training epoch
  2. Can avoid getting caught up in local optima or saddle points

Is it possible to say that all training can depend on the pretraining model？

Is there a downside to loading the pre-training model？