Added capabilities to load local models

From https://stackoverflow.com/questions/75886674/how-to-compute-sentence-level-perplexity-from-hugging-face-language-models/75887046?noredirect=1#comment133854598_75887046

perplexity.py

@@ -101,7 +101,9 @@ class Perplexity(evaluate.Metric):
         )
 
     def _compute(
-        self, predictions, model_id, batch_size: int = 16, add_start_token: bool = True, device=None, max_length=None
+        self, predictions, model_id, batch_size: int = 16, add_start_token: bool = True, local_file_only: bool = False,
+        device=None, max_length=None,
+        
     ):
 
         if device is not None:
@@ -111,10 +113,10 @@ class Perplexity(evaluate.Metric):
         else:
             device = "cuda" if torch.cuda.is_available() else "cpu"
 
-        model = AutoModelForCausalLM.from_pretrained(model_id)
+        model = AutoModelForCausalLM.from_pretrained(model_id, local_file_only=local_file_only)
         model = model.to(device)
 
-        tokenizer = AutoTokenizer.from_pretrained(model_id)
+        tokenizer = AutoTokenizer.from_pretrained(model_id, local_file_only=local_file_only)
 
         # if batch_size > 1 (which generally leads to padding being required), and
         # if there is not an already assigned pad_token, assign an existing