diff --git a/README.md b/README.md
index 6d24eaf..9bfaeeb 100644
--- a/README.md
+++ b/README.md
@@ -112,7 +112,7 @@ import torch
 from transformers import AutoModelForCausalLM
 
 from deepseek_vl.models import VLChatProcessor, MultiModalityCausalLM
-from deepseek_vl.utils.io import load_pil_images
+from deepseek_vl.utils.io import load_pil_images, get_device_and_dtype
 
 
 # specify the path to the model
@@ -121,7 +121,9 @@ vl_chat_processor: VLChatProcessor = VLChatProcessor.from_pretrained(model_path)
 tokenizer = vl_chat_processor.tokenizer
 
 vl_gpt: MultiModalityCausalLM = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True)
-vl_gpt = vl_gpt.to(torch.bfloat16).cuda().eval()
+
+device, dtype = get_device_and_dtype()
+vl_gpt = vl_gpt.to(dtype).to(device).eval()
 
 conversation = [
     {