Enhance model loading logic in chatbot application to support direct loading and PEFT adapter fallback. Updated model and tokenizer initialization for improved error handling and device management.

app.py

@@ -1,27 +1,40 @@
 import gradio as gr
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
 
 
 # Load tokenizer và model
 print("Đang tải model...")
-model_name = "cochi1706/decoder"
-subfolder = "qwen3-finetuned"
+
+# Thử load trực tiếp từ path đầy đủ trước
+try:
+    model_name = "cochi1706/decoder/qwen3-finetuned"
+    print(f"Đang thử load model từ: {model_name}")
+    tokenizer = AutoTokenizer.from_pretrained(model_name)
+    model = AutoModelForCausalLM.from_pretrained(
+        model_name,
+        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+        device_map="auto" if torch.cuda.is_available() else None,
+    )
+except Exception as e:
+    print(f"Không thể load trực tiếp: {e}")
+    print("Đang thử load như PEFT adapter...")
+    # Nếu không được, thử load như PEFT adapter
+    base_model_name = "Qwen/Qwen3-0.6B"
+    adapter_repo = "cochi1706/coding-assistant"
+    
+    base_model = AutoModelForCausalLM.from_pretrained(
+        base_model_name,
+        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+        device_map="auto" if torch.cuda.is_available() else None,
+    )
+    model = PeftModel.from_pretrained(base_model, adapter_repo)
+    tokenizer = AutoTokenizer.from_pretrained(adapter_repo)
 
 # Xác định device
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
-# Load tokenizer
-tokenizer = AutoTokenizer.from_pretrained(model_name, subfolder=subfolder)
-
-# Load model
-model = AutoModelForCausalLM.from_pretrained(
-    model_name,
-    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
-    device_map="auto" if torch.cuda.is_available() else None,
-    subfolder=subfolder,
-)
-
 # Set padding token nếu chưa có
 if tokenizer.pad_token is None:
     tokenizer.pad_token = tokenizer.eos_token