removed to.cuda in 4bit

handler.py

@@ -27,9 +27,10 @@ class EndpointHandler():
         self.model = AutoModelForCausalLM.from_pretrained(
             path,
             temperature=0,
-            torch_dtype = self.dtype,
+#            torch_dtype = self.dtype,
             load_in_4bit = self.load_in_4bit,
-        ).to("cuda")
+        )
+#        .to("cuda")
 
         self.tokenizer = AutoTokenizer.from_pretrained(path)
         self.tokenizer.padding_side="left"