Update model_setup.py

model_setup.py

@@ -19,15 +19,16 @@ def load_model():
     token = os.getenv("HF_TOKEN")  # Lấy token từ biến môi trường
 
 
-    config = PeftConfig.from_pretrained("anhvv200053/Vinallama-2-7B-updated1-instruction-v2")
-    base_model = AutoModelForCausalLM.from_pretrained("vilm/vinallama-2.7b-chat")
-    model = PeftModel.from_pretrained(base_model, "anhvv200053/Vinallama-2-7B-updated1-instruction-v2", token = token)
+    # config = PeftConfig.from_pretrained("anhvv200053/Vinallama-2-7B-updated1-instruction-v2")
+    # base_model = AutoModelForCausalLM.from_pretrained("vilm/vinallama-2.7b-chat")
+    # model = PeftModel.from_pretrained(base_model, "anhvv200053/Vinallama-2-7B-updated1-instruction-v2", token = token)
     
-    # model = AutoModelForCausalLM.from_pretrained(
-    #     "anhvv200053/Vinallama-2-7B-updated1-instruction-v2",
-    #     device_map={"": 0},
-    #     token = token
-    # )
+    model = AutoModelForCausalLM.from_pretrained(
+        "anhvv200053/Vinallama-2-7B-updated1-instruction-v2",
+        quantization_config=bnb_config,
+        device_map={"": "cpu"},  # Đặt sử dụng CPU
+        token = token
+    )
     model.config.pretraining_tp = 1
 
     tokenizer = AutoTokenizer.from_pretrained('anhvv200053/Vinallama-2-7B-updated1-instruction-v2', trust_remote_code=True, use_fast=True, token = token)