Model evaluation inconsistency between runs

[xinmengZ]

I ran into inconsistent evaluation results while using TxGNN_Demo.ipynb. When I executed the evaluation immediately after training:

TxGNN.finetune(n_epoch = 500, 
               learning_rate = 5e-4,
               train_print_per_n = 5,
               valid_per_n = 20)
result = TxEval.eval_disease_centric(disease_idxs = 'test_set', 
                                     show_plot = False, 
                                     verbose = True, 
                                     save_result = True,
                                     return_raw = False)

I obtained different results compared to when I first saved and then reloaded the model:

TxGNN.finetune(n_epoch = 500, 
               learning_rate = 5e-4,
               train_print_per_n = 5,
               valid_per_n = 20)
TxGNN.save_model('./model_ckpt')
TxGNN.load_pretrained('./model_ckpt')
result = TxEval.eval_disease_centric(disease_idxs = 'test_set', 
                                     show_plot = False, 
                                     verbose = True, 
                                     save_result = True,
                                     return_raw = False)

Did I use load_pretrained() incorrectly, or is this an issue others have encountered as well?