Upload handler.py

handler.py

@@ -32,12 +32,12 @@ class EndpointHandler():
 
         self.model = AutoModel.from_pretrained(
             model_dir,
-            quantization_config=bnb_config,
+            #quantization_config=bnb_config,
             device_map="auto",
             trust_remote_code=True,
             attn_implementation="eager",
         )
-        self.model = PeftModel.from_pretrained(self.model, model_dir, subfolder='lora')
+        #self.model = PeftModel.from_pretrained(self.model, model_dir, subfolder='lora')
         self.model.eval()