Update app.py

app.py

@@ -183,7 +183,7 @@ def on_btn_click():
 def load_model():
     model = (AutoModelForCausalLM.from_pretrained(
         model_name_or_path,
-        trust_remote_code=True).to(torch.float32).cuda())
+        trust_remote_code=True).to(torch.float32).cuda()).half()
     tokenizer = AutoTokenizer.from_pretrained(model_name_or_path,
                                               trust_remote_code=True)
     return model, tokenizer