Update inference.py

inference.py

@@ -40,7 +40,7 @@ class DualBertModel: # use this class to encode your document and queries.
         attention_mask = torch.tensor(encoded.get('attention_mask')).to(device)
         return input_ids, attention_mask
     
-    def encode_queries(self, queries, batch_size: int, **kwargs) -> np.ndarray:
+    def encode_queries(self, queries, batch_size: int, **kwargs):
         """
         Encodes a list of strings (queries) and returns a list of encodings.
         """
@@ -54,7 +54,7 @@ class DualBertModel: # use this class to encode your document and queries.
         return torch.concat(to_return)
             
     
-    def encode_corpus(self, corpus, batch_size: int, **kwargs) -> np.ndarray:
+    def encode_corpus(self, corpus, batch_size: int, **kwargs):
         """
         Encodes a list of strings (documents) and returns a list of encodings.
         """