Update handler.py

handler.py

@@ -4,12 +4,12 @@ from typing import Dict, List, Any
 from llama_cpp import Llama
 import gemma_tools as gem
 
-MAX_TOKENS = 512
+MAX_TOKENS = 8192
 
 class EndpointHandler():
     def __init__(self, data):
-        # Update the model path and filename with your ComicBot model
-        self.model = Llama.from_pretrained("njwright92/ComicBot_v.2-gguf", filename="ComicBot_v.2-q4_k_m.gguf", n_ctx=8192)
+        # Updated the model path and filename with your ComicBot model
+        self.model = Llama.from_pretrained("njwright92/ComicBot_v.2-gguf", filename="comic_mistral-v5.2.q5_0.gguf", n_ctx=8192)
 
     def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
         args = gem.get_args_or_none(data)
@@ -40,3 +40,4 @@ class EndpointHandler():
         res = self.model(fmat, temperature=args["temperature"], top_p=args["top_p"], top_k=args["top_k"], max_tokens=max_length)
 
         return res
+