Update input ids filtering

handler.py

@@ -71,9 +71,9 @@ class EndpointHandler:
         if parameters is not None:
             outputs = self.model.generate(input_ids=input_ids, **parameters)
         else:
-            outputs = self.model.generate(input_ids=input_ids, max_new_tokens=64)
+            outputs = self.model.generate(input_ids=input_ids, max_new_tokens=20)
 
         # postprocess the prediction
-        prediction = self.tokenizer.decode(outputs[0]) #, skip_special_tokens=True)
+        prediction = self.tokenizer.decode(outputs[0][input_ids.shape[1]:]) #, skip_special_tokens=True)
         prediction = prediction.split("<|endoftext|>")[0]
         return [{"generated_text": prediction}]