Hugging Face's logo

bytedance-research
/
UI-TARS-7B-SFT

Image-Text-to-Text
Transformers
Safetensors
English
qwen2_vl
multimodal
gui
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
2
nielsr HF staff commited on
Commit
d97fe88
·
verified ·
1 Parent(s): 23b6e9c

Add link to paper

Browse files

This PR ensures the model can be found at https://huggingface.co./papers/2501.12326.

Files changed (1) hide show
  1. README.md +3 -78
README.md CHANGED
@@ -9,7 +9,6 @@ tags:
9
  library_name: transformers
10
  ---
11
 
12
-
13
  # UI-TARS-7B-SFT
14
  [UI-TARS-2B-SFT](https://huggingface.co/bytedance-research/UI-TARS-2B-SFT)  | 
15
  [UI-TARS-7B-SFT](https://huggingface.co/bytedance-research/UI-TARS-7B-SFT)  | 
@@ -19,7 +18,9 @@ library_name: transformers
19
  ## Introduction
20
 
21
  UI-TARS is a next-generation native GUI agent model designed to interact seamlessly with graphical user interfaces (GUIs) using human-like perception, reasoning, and action capabilities. Unlike traditional modular frameworks, UI-TARS integrates all key components—perception, reasoning, grounding, and memory—within a single vision-language model (VLM), enabling end-to-end task automation without predefined workflows or manual rules.
22
- <!-- ![Local Image](figures/UI-TARS.png) -->
 
 
23
  <p align="center">
24
  <img src="https://github.com/bytedance/UI-TARS/blob/main/figures/UI-TARS-vs-Previous-SOTA.png?raw=true" width="90%"/>
25
  <p>
@@ -27,10 +28,6 @@ UI-TARS is a next-generation native GUI agent model designed to interact seamles
27
  <img src="https://github.com/bytedance/UI-TARS/blob/main/figures/UI-TARS.png?raw=true" width="90%"/>
28
  <p>
29
 
30
- <!-- ![Local Image](figures/UI-TARS-vs-Previous-SOTA.png) -->
31
-
32
- This repository contains the model for the paper [UI-TARS: Pioneering Automated GUI Interaction with Native Agents](https://huggingface.co/papers/2501.12326).
33
-
34
  Code: https://github.com/bytedance/UI-TARS
35
 
36
  ## Performance
@@ -69,35 +66,6 @@ Code: https://github.com/bytedance/UI-TARS
69
  | **UI-TARS-72B** | **63.0** | **17.3** | **40.8** | **57.1** | **15.4** | **39.6** | 18.8 | **12.5**| 17.2 | **64.6** | 20.9 | 45.7 | **63.3** | **26.4** | **54.8** | **42.1**| 15.7 | **30.1**| **50.9**| **17.5**| **38.1** |
70
 
71
 
72
- - **ScreenSpot**
73
-
74
- | Method | Mobile-Text | Mobile-Icon/Widget | Desktop-Text | Desktop-Icon/Widget | Web-Text | Web-Icon/Widget | Avg |
75
- |--------|-------------|-------------|-------------|-------------|-------------|---------|---------|
76
- | **Agent Framework** | | | | | | | |
77
- | GPT-4 (SeeClick) | 76.6 | 55.5 | 68.0 | 28.6 | 40.9 | 23.3 | **48.8** |
78
- | GPT-4 (OmniParser) | 93.9 | 57.0 | 91.3 | 63.6 | 81.3 | 51.0 | **73.0** |
79
- | GPT-4 (UGround-7B) | 90.1 | 70.3 | 87.1 | 55.7 | 85.7 | 64.6 | **75.6** |
80
- | GPT-4o (SeeClick) | 81.0 | 59.8 | 69.6 | 33.6 | 43.9 | 26.2 | **52.3** |
81
- | GPT-4o (UGround-7B) | 93.4 | 76.9 | 92.8 | 67.9 | 88.7 | 68.9 | **81.4** |
82
- | **Agent Model** | | | | | | | |
83
- | GPT-4 | 22.6 | 24.5 | 20.2 | 11.8 | 9.2 | 8.8 | **16.2** |
84
- | GPT-4o | 20.2 | 24.9 | 21.1 | 23.6 | 12.2 | 7.8 | **18.3** |
85
- | CogAgent | 67.0 | 24.0 | 74.2 | 20.0 | 70.4 | 28.6 | **47.4** |
86
- | SeeClick | 78.0 | 52.0 | 72.2 | 30.0 | 55.7 | 32.5 | **53.4** |
87
- | Qwen2-VL | 75.5 | 60.7 | 76.3 | 54.3 | 35.2 | 25.7 | **55.3** |
88
- | UGround-7B | 82.8 | 60.3 | 82.5 | 63.6 | 80.4 | 70.4 | **73.3** |
89
- | Aguvis-G-7B | 88.3 | 78.2 | 88.1 | 70.7 | 85.7 | 74.8 | **81.8** |
90
- | OS-Atlas-7B | 93.0 | 72.9 | 91.8 | 62.9 | 90.9 | 74.3 | **82.5** |
91
- | Claude Computer Use | - | - | - | - | - | - | **83.0** |
92
- | Gemini 2.0 (Project Mariner) | - | - | - | - | - | - | **84.0** |
93
- | Aguvis-7B | **95.6** | 77.7 | 93.8 | 67.1 | 88.3 | 75.2 | **84.4** |
94
- | Aguvis-72B | 94.5 | **85.2** | 95.4 | 77.9 | **91.3** | **85.9** | **89.2** |
95
- | **Our Model** | | | | | | | |
96
- | **UI-TARS-2B** | 93.0 | 75.5 | 90.7 | 68.6 | 84.3 | 74.8 | **82.3** |
97
- | **UI-TARS-7B** | 94.5 | **85.2** | **95.9** | 85.7 | 90.0 | 83.5 | **89.5** |
98
- | **UI-TARS-72B** | 94.9 | 82.5 | 89.7 | **88.6** | 88.7 | 85.0 | **88.4** |
99
-
100
-
101
  - **ScreenSpot v2**
102
 
103
  | Method | Mobile-Text | Mobile-Icon/Widget | Desktop-Text | Desktop-Icon/Widget | Web-Text | Web-Icon/Widget | Avg |
@@ -116,49 +84,6 @@ Code: https://github.com/bytedance/UI-TARS
116
  | **UI-TARS-72B** | 94.8 | 86.3 | 91.2 | **87.9** | 91.5 | **87.7** | **90.3** |
117
 
118
 
119
- **Offline Agent Capability Evaluation**
120
- - **Multimodal Mind2Web**
121
-
122
- | Method | Cross-Task Ele.Acc | Cross-Task Op.F1 | Cross-Task Step SR | Cross-Website Ele.Acc | Cross-Website Op.F1 | Cross-Website Step SR | Cross-Domain Ele.Acc | Cross-Domain Op.F1 | Cross-Domain Step SR |
123
- |--------|----------------------|-------------------|--------------------|----------------------|--------------------|-------------------|--------------------|-------------------|-------------------|
124
- | **Agent Framework** | | | | | | | | | |
125
- | GPT-4o (SeeClick) | 32.1 | - | - | 33.1 | - | - | 33.5 | - | - |
126
- | GPT-4o (UGround) | 47.7 | - | - | 46.0 | - | - | 46.6 | - | - |
127
- | GPT-4o (Aria-UI) | 57.6 | - | - | 57.7 | - | - | 61.4 | - | - |
128
- | GPT-4V (OmniParser) | 42.4 | 87.6 | 39.4 | 41.0 | 84.8 | 36.5 | 45.5 | 85.7 | 42.0 |
129
- | **Agent Model** | | | | | | | | | |
130
- | GPT-4o | 5.7 | 77.2 | 4.3 | 5.7 | 79.0 | 3.9 | 5.5 | 86.4 | 4.5 |
131
- | GPT-4 (SOM) | 29.6 | - | 20.3 | 20.1 | - | 13.9 | 27.0 | - | 23.7 |
132
- | GPT-3.5 (Text-only) | 19.4 | 59.2 | 16.8 | 14.9 | 56.5 | 14.1 | 25.2 | 57.9 | 24.1 |
133
- | GPT-4 (Text-only) | 40.8 | 63.1 | 32.3 | 30.2 | 61.0 | 27.0 | 35.4 | 61.9 | 29.7 |
134
- | Claude | 62.7 | 84.7 | 53.5 | 59.5 | 79.6 | 47.7 | 64.5 | 85.4 | 56.4 |
135
- | Aguvis-7B | 64.2 | 89.8 | 60.4 | 60.7 | 88.1 | 54.6 | 60.4 | 89.2 | 56.6 |
136
- | CogAgent | - | - | 62.3 | - | - | 54.0 | - | - | 59.4 |
137
- | Aguvis-72B | 69.5 | 90.8 | 64.0 | 62.6 | 88.6 | 56.5 | 63.5 | 88.5 | 58.2 |
138
- | **Our Model** | | | | | | | | | |
139
- | **UI-TARS-2B** | 62.3 | 90.0 | 56.3 | 58.5 | 87.2 | 50.8 | 58.8 | 89.6 | 52.3 |
140
- | **UI-TARS-7B** | 73.1 | 92.2 | 67.1 | 68.2 | 90.9 | 61.7 | 66.6 | 90.9 | 60.5 |
141
- | **UI-TARS-72B** | **74.7** | **92.5** | **68.6** | **72.4** | **91.2** | **63.5** | **68.9** | **91.8** | **62.1** |
142
-
143
-
144
- - **Android Control and GUI Odyssey**
145
-
146
- | Agent Models | AndroidControl-Low Type | AndroidControl-Low Grounding | AndroidControl-Low SR | AndroidControl-High Type | AndroidControl-High Grounding | AndroidControl-High SR | GUIOdyssey Type | GUIOdyssey Grounding | GUIOdyssey SR |
147
- |---------------------|----------------------|----------------------|----------------|----------------------|----------------------|----------------|----------------|----------------|----------------|
148
- | Claude | 74.3 | 0.0 | 19.4 | 63.7 | 0.0 | 12.5 | 60.9 | 0.0 | 3.1 |
149
- | GPT-4o | 74.3 | 0.0 | 19.4 | 66.3 | 0.0 | 20.8 | 34.3 | 0.0 | 3.3 |
150
- | SeeClick | 93.0 | 73.4 | 75.0 | 82.9 | 62.9 | 59.1 | 71.0 | 52.4 | 53.9 |
151
- | InternVL-2-4B | 90.9 | 84.1 | 80.1 | 84.1 | 72.7 | 66.7 | 82.1 | 55.5 | 51.5 |
152
- | Qwen2-VL-7B | 91.9 | 86.5 | 82.6 | 83.8 | 77.7 | 69.7 | 83.5 | 65.9 | 60.2 |
153
- | Aria-UI | -- | 87.7 | 67.3 | -- | 43.2 | 10.2 | -- | 86.8 | 36.5 |
154
- | OS-Atlas-4B | 91.9 | 83.8 | 80.6 | 84.7 | 73.8 | 67.5 | 83.5 | 61.4 | 56.4 |
155
- | OS-Atlas-7B | 93.6 | 88.0 | 85.2 | 85.2 | 78.5 | 71.2 | 84.5 | 67.8 | 62.0 |
156
- | Aguvis-7B | -- | -- | 80.5 | -- | -- | 61.5 | -- | -- | -- |
157
- | Aguvis-72B | -- | -- | 84.4 | -- | -- | 66.4 | -- | -- | -- |
158
- | **UI-TARS-2B** | **98.1** | 87.3 | 89.3 | 81.2 | 78.4 | 68.9 | 93.9 | 86.8 | 83.4 |
159
- | **UI-TARS-7B** | 98.0 | 89.3 | 90.8 | 83.7 | 80.5 | 72.5 | 94.6 | 90.1 | 87.0 |
160
- | **UI-TARS-72B** | **98.1** | **89.9** | **91.3** | **85.2** | **81.5** | **74.7** | **95.4** | **91.4** | **88.6** |
161
-
162
  **Online Agent Capability Evaluation**
163
 
164
  | Method | OSWorld (Online) | AndroidWorld (Online) |
 
9
  library_name: transformers
10
  ---
11
 
 
12
  # UI-TARS-7B-SFT
13
  [UI-TARS-2B-SFT](https://huggingface.co/bytedance-research/UI-TARS-2B-SFT) &nbsp;|&nbsp;
14
  [UI-TARS-7B-SFT](https://huggingface.co/bytedance-research/UI-TARS-7B-SFT) &nbsp;|&nbsp;
 
18
  ## Introduction
19
 
20
  UI-TARS is a next-generation native GUI agent model designed to interact seamlessly with graphical user interfaces (GUIs) using human-like perception, reasoning, and action capabilities. Unlike traditional modular frameworks, UI-TARS integrates all key components—perception, reasoning, grounding, and memory—within a single vision-language model (VLM), enabling end-to-end task automation without predefined workflows or manual rules.
21
+
22
+ This repository contains the model for the paper [UI-TARS: Pioneering Automated GUI Interaction with Native Agents](https://huggingface.co/papers/2501.12326).
23
+
24
  <p align="center">
25
  <img src="https://github.com/bytedance/UI-TARS/blob/main/figures/UI-TARS-vs-Previous-SOTA.png?raw=true" width="90%"/>
26
  <p>
 
28
  <img src="https://github.com/bytedance/UI-TARS/blob/main/figures/UI-TARS.png?raw=true" width="90%"/>
29
  <p>
30
 
 
 
 
 
31
  Code: https://github.com/bytedance/UI-TARS
32
 
33
  ## Performance
 
66
  | **UI-TARS-72B** | **63.0** | **17.3** | **40.8** | **57.1** | **15.4** | **39.6** | 18.8 | **12.5**| 17.2 | **64.6** | 20.9 | 45.7 | **63.3** | **26.4** | **54.8** | **42.1**| 15.7 | **30.1**| **50.9**| **17.5**| **38.1** |
67
 
68
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
69
  - **ScreenSpot v2**
70
 
71
  | Method | Mobile-Text | Mobile-Icon/Widget | Desktop-Text | Desktop-Icon/Widget | Web-Text | Web-Icon/Widget | Avg |
 
84
  | **UI-TARS-72B** | 94.8 | 86.3 | 91.2 | **87.9** | 91.5 | **87.7** | **90.3** |
85
 
86
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
87
  **Online Agent Capability Evaluation**
88
 
89
  | Method | OSWorld (Online) | AndroidWorld (Online) |