fix error about max_length

handler.py

@@ -34,7 +34,6 @@ class EndpointHandler:
             path, device_map="auto", torch_dtype=torch.bfloat16
         )
         LOGGER.info(f"Inference model loaded from {path}")
-        LOGGER.info(f"Model outline: {self.model}")
         LOGGER.info(f"Model device: {self.model.device}")
 
         # Fix the pad and bos tokens to avoid bug in the tokenizer
@@ -45,7 +44,7 @@ class EndpointHandler:
         )
 
     def check_valid_inputs(
-        self, input_a: str, input_b: str, task: int, parameters: Dict[str, Any]
+        self, input_a: str, input_b: str, task: int
     ) -> bool:
         """
         Check if the inputs are valid
@@ -104,11 +103,13 @@ class EndpointHandler:
         input_a = data.pop("input_a", None)
         input_b = data.pop("input_b", None)
         task = data.pop("task", None)
-        parameters = data.pop("parameters", None)
+        parameters = data.pop("parameters", {})
 
         # Check valid inputs
-        if not self.check_valid_inputs(input_a, input_b, task, parameters):
-            return []
+        if not self.check_valid_inputs(input_a, input_b, task):
+            return [{"error": "Invalid inputs"}]
+        if "max_new_tokens" not in parameters:
+            parameters["max_new_tokens"] = 512
 
         # Tokenize the input
         tokenized_input = self.tokenize_input(input_a, input_b, task)