Update handler.py

handler.py

@@ -10,7 +10,7 @@ MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "8192"))
 
 class EndpointHandler:
     def __init__(self, path=""):
-        self.model_name_or_path = "ClosedCharacter/Peach-9B-8k-Roleplay"
+        self.model_name_or_path = "threadshare/Peach-9B-8k-Roleplay"
         self.tokenizer = AutoTokenizer.from_pretrained(self.model_name_or_path, use_fast=True, flash_atten=True)
         self.model = AutoModelForCausalLM.from_pretrained(
             self.model_name_or_path, torch_dtype=torch.bfloat16,