Move paht to a variable

demo_rag.py

@@ -4,10 +4,11 @@ import torch
 import time
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+pretrain_path = "./deepseek-coder-1.3b-instruct"
 
 # Load tokenizer and model
-tokenizer = AutoTokenizer.from_pretrained("./deepseek-coder-1.3b-instruct")
-model = AutoModelForCausalLM.from_pretrained("./deepseek-coder-1.3b-instruct", torch_dtype=torch.bfloat16, device_map=device)
+tokenizer = AutoTokenizer.from_pretrained(pretrain_path)
+model = AutoModelForCausalLM.from_pretrained(pretrain_path, torch_dtype=torch.bfloat16, device_map=device)
 
 # Initialize RAG and add schema docs
 retriever = SQLMetadataRetriever()