Update model.py

model.py

@@ -18,7 +18,13 @@ if torch.cuda.is_available():
         use_safetensors=False,
     )
 else:
-    model = None
+    # Switch PyTorch to CPU mode
+    torch.set_default_tensor_type(torch.FloatTensor)
+    model = AutoModelForCausalLM.from_pretrained(
+        model_id, 
+        torch_dtype=torch.float32,
+        device_map='cpu') 
+    
 tokenizer = AutoTokenizer.from_pretrained(model_id)