Update app.py

app.py

@@ -66,14 +66,13 @@ transformers.utils.logging.enable_explicit_format()
 model_id = "microsoft/bitnet-b1.58-2B-4T-bf16"
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 hf_save_dir = "./bitnet"
-
+model = AutoModelForCausalLM.from_pretrained(model_id)
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 if torch.cuda.is_available():
     print("CUDA is available. Using GPU:", torch.cuda.get_device_name(0))
-    model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16)
 else:
     print("CUDA not available; using CPU.")
-    model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float32)
+
 
 model.to(device)
 print(f"Loaded pre-trained Hugging Face model '{model_id}'.")