updates

app.py

@@ -5,16 +5,15 @@ import torch
 from PIL import Image, ImageDraw
 import gradio as gr
 
-from transformers import Qwen2VLModel
-
-from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
+from transformers import Qwen2_5_VLForConditionalGeneration
+from transformers import AutoProcessor
 from qwen_vl_utils import process_vision_info  # include this file in your repo if not pip-installable
 
 # ---- model & processor loaded on CPU ----
-model = Qwen2VLModel.from_pretrained(
+model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     "ByteDance-Seed/UI-TARS-1.5-7B",
     device_map="auto",
-    torch_dtype=torch.float16,    # CPU-friendly
+    torch_dtype=torch.float32,    # CPU-friendly
 )
 processor = AutoProcessor.from_pretrained(
     "ByteDance-Seed/UI-TARS-1.5-7B",

requirements.txt

@@ -1,4 +1,4 @@
-transformers
+git+https://github.com/huggingface/transformers.git@9985d06add07a4cc691dc54a7e34f54205c04d40
 accelerate
 torch
 torchvision