Update perplexity.py

Add model & tokenizer parameters, to avoid reinitialising the model every time on _compute()

perplexity.py

@@ -49,6 +49,8 @@ Args:
     add_start_token (bool): whether to add the start token to the texts,
         so the perplexity can include the probability of the first word. Defaults to True.
     device (str): device to run on, defaults to 'cuda' when available
+    model (AutoModelForCausalLM): the model for calculating Perplexity, if provided, the model won't initialized from model_id
+    tokenizer (AutoTokenizer): the tokenizer for calculating Perplexity, if provided, the tokenizer won't initialized from model_id
 Returns:
     perplexity: dictionary containing the perplexity scores for the texts
         in the input list, as well as the mean perplexity. If one of the input texts is
@@ -101,7 +103,7 @@ 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, device=None, max_length=None, model = None, tokenizer = None
     ):
 
         if device is not None:
@@ -110,11 +112,11 @@ class Perplexity(evaluate.Metric):
                 device = "cuda"
         else:
             device = "cuda" if torch.cuda.is_available() else "cpu"
-
-        model = AutoModelForCausalLM.from_pretrained(model_id)
-        model = model.to(device)
-
-        tokenizer = AutoTokenizer.from_pretrained(model_id)
+        if model is None:
+            model = AutoModelForCausalLM.from_pretrained(model_id)
+            model = model.to(device)
+        if tokenizer is None:
+            tokenizer = AutoTokenizer.from_pretrained(model_id)
 
         # if batch_size > 1 (which generally leads to padding being required), and
         # if there is not an already assigned pad_token, assign an existing