Update utils.py

utils.py

@@ -83,8 +83,8 @@ def load_tokenizer_and_model(base_model,load_8bit=False):
     if device == "cuda":
         model = AutoModelForCausalLM.from_pretrained(
             base_model,
-            load_in_8bit=load_8bit,
-            torch_dtype=torch.float16,
+            #load_in_8bit=load_8bit,
+            #torch_dtype=torch.float16,
             device_map="auto",
         )
     else:
@@ -92,8 +92,8 @@ def load_tokenizer_and_model(base_model,load_8bit=False):
             base_model, device_map={"": device}, low_cpu_mem_usage=True
         )
 
-    if not load_8bit:
-        model.half()  # seems to fix bugs for some users.
+    #if not load_8bit:
+        #model.half()  # seems to fix bugs for some users.
 
     model.eval()
     return tokenizer,model,device