update

Author: yuanlehome

llm/predict/predictor.py

@@ -135,16 +135,15 @@ class PredictorArgument:
     )
 
     total_max_length: int = field(
-        default=8192, metadata={"help": "Super parameter. Maximum sequence length(encoder+decoder)."}
+        default=2048, metadata={"help": "Super parameter. Maximum sequence length(encoder+decoder)."}
     )
 
-    def __post_init__(self):
-        if self.device == "npu":
-            self.total_max_length = self.src_length + self.max_length
-
     def __post_init__(self):
         if self.append_attn:
             self.block_attn = True
+        assert (
+            self.src_length + self.max_length <= self.total_max_length
+        ), "src_length + max_length should smaller than total_max_length."
 
 
 @dataclass