Create app.py

app.py

@@ -0,0 +1,5 @@
+# Load model directly
+from transformers import AutoProcessor, AutoModelForImageTextToText
+
+processor = AutoProcessor.from_pretrained("bytedance-research/UI-TARS-2B-SFT")
+model = AutoModelForImageTextToText.from_pretrained("bytedance-research/UI-TARS-2B-SFT")