Set half revision from base model weights

app.py

@@ -10,6 +10,8 @@ model = AutoModelForCausalLM.from_pretrained(
     return_dict=True,
     load_in_8bit=True,
     device_map="auto",
+    revision="half",
+    # low_cpu_mem_usage=True
 )
 tokenizer = AutoTokenizer.from_pretrained(peft_model_id)
 # Load the Lora model