Fix bug of low_cpu_mem_usage in finetune.py.

finetune.py

@@ -291,9 +291,9 @@ def train():
     ):
         raise RuntimeError("ZeRO3 is incompatible with LoRA when finetuning on base model.")
 
-    model_load_kwargs = {}
+    model_load_kwargs = {
-    if deepspeed.is_deepspeed_zero3_enabled():
+        'low_cpu_mem_usage': not deepspeed.is_deepspeed_zero3_enabled(),
-        model_load_kwargs['low_cpu_mem_usage'] = False
+    }
 
     # Set RoPE scaling factor
     config = transformers.AutoConfig.from_pretrained(