Load fine-tuned model

Hello, I have a file where i trained a BERT model and saved the state_dict of the model. How can i load the model in another .py file and use it to predict the polarity some keyboard input sentences? I have the below code :

 import torch
from transformers import BertTokenizer

tokenizer = BertTokenizer.from_pretrained('bert-base-multilingual-uncased', do_lower_case=True)
inputs = tokenizer("I do not feel good today.", return_tensors="pt")
outputs = model(**inputs)

Assuming that I do not feel good today. is the sentence entered from keyboard, i would like the model to return 0 if it is negative or 1 if it is positive. How can I define the model so it will not be linked to the py file where I trained the model? I tried the below but that will start the training again( BertClassifier is the class where i trained the model) :

from bert import BertClassifier 
model = BertClassifier(freeze_bert=False)