Update handler.py

handler.py

@@ -51,7 +51,7 @@ class EndpointHandler:
                 return_tensors="pt",
                 padding=True,
                 truncation=True,
-                max_length=1026
+                max_length=1024
             ).to(self.device)
             
             # Génération
@@ -59,7 +59,7 @@ class EndpointHandler:
                 generated_tokens = self.model.generate(
                     **inputs,
                     forced_bos_token_id=self.tokenizer.lang_code_to_id["en_XX"],
-                    max_length=1026,
+                    max_length=1024,
                     num_beams=4,
                     length_penalty=1.0,
                     do_sample=False