Update handler.py

handler.py

@@ -3,7 +3,7 @@ from typing import Dict, List, Any
 from llama_cpp import Llama
 import gemma_tools
 
-MAX_TOKENS = 5000
+MAX_TOKENS = 1000
 
 
 class EndpointHandler():
@@ -51,7 +51,7 @@ class EndpointHandler():
                 "detail": str(e)
             }]
 
-        max_length = data.get("max_length", 512)
+        max_length = data.get("max_length", 212)
         try:
             max_length = int(max_length)
             print(f"Max length set to: {max_length}")