Hugging Face's logo

lgy0404
/
MemGUI-8B-SFT

Image-Text-to-Text
Transformers
Safetensors
English
qwen3_vl
gui-agent
mobile-gui
android
memory
context-management
conact
memgui-agent
long-horizon
conversational
Eval Results (legacy)
Model card Files
xet
 Community
1

Add links to paper, code and project page

#1
by nielsr HF Staff - opened
base: refs/heads/main
←
from: refs/pr/1
Discussion Files changed
+55
-54
Files changed (1) hide show
  1. README.md +55 -54
README.md CHANGED
@@ -1,58 +1,57 @@
1
  ---
2
- license: apache-2.0
 
 
3
  language:
4
- - en
5
  library_name: transformers
 
6
  pipeline_tag: image-text-to-text
7
- base_model: Qwen/Qwen3-VL-8B-Instruct
8
- datasets:
9
- - lgy0404/MemGUI-3K
10
  tags:
11
- - qwen3_vl
12
- - gui-agent
13
- - mobile-gui
14
- - android
15
- - memory
16
- - context-management
17
- - conact
18
- - memgui-agent
19
- - long-horizon
20
  model-index:
21
- - name: MemGUI-8B-SFT
22
- results:
23
- - task:
24
- type: image-text-to-text
25
- name: Long-horizon mobile GUI control
26
- dataset:
27
- type: lgy0404/MemGUI-3K
28
- name: MemGUI-Bench
29
- metrics:
30
- - type: pass_at_1
31
- name: Pass@1
32
- value: 23.4
33
- unit: "%"
34
- - type: pass_at_3
35
- name: Pass@3
36
- value: 35.9
37
- unit: "%"
38
- - type: irr
39
- name: Information Retention Rate
40
- value: 30.2
41
- unit: "%"
42
- - task:
43
- type: image-text-to-text
44
- name: Out-of-distribution mobile GUI control
45
- dataset:
46
- type: mobileworld-gui-only
47
- name: MobileWorld GUI-Only
48
- metrics:
49
- - type: success_rate
50
- name: Success Rate
51
- value: 17.9
52
- unit: "%"
53
  ---
 
54
  # MemGUI-8B-SFT
55
 
 
 
56
  MemGUI-8B-SFT is an 8B MemGUI-Agent model trained from Qwen3-VL-8B-Instruct on
57
  MemGUI-3K. It is designed for long-horizon mobile GUI control with proactive
58
  context management.
@@ -65,12 +64,12 @@ Folded UI State, and Recent Step Record.
65
 
66
  ## Model Details
67
 
68
- - Model type: multimodal mobile GUI agent
69
- - Base model: `Qwen/Qwen3-VL-8B-Instruct`
70
- - Training data: `lgy0404/MemGUI-3K`
71
- - Training recipe: supervised fine-tuning with ms-swift
72
- - Output protocol: ConAct 5-part structured output
73
- - License: Apache 2.0
74
 
75
  ## Intended Use
76
 
@@ -131,6 +130,8 @@ Dataset page: https://huggingface.co/datasets/lgy0404/MemGUI-3K
131
  ```bibtex
132
  @article{memguiagent2026,
133
  title = {MemGUI-Agent: An End-to-End Long-Horizon Mobile GUI Agent with Proactive Context Management},
134
- year = {2026}
 
 
135
  }
136
- ```
 
1
  ---
2
+ base_model: Qwen/Qwen3-VL-8B-Instruct
3
+ datasets:
4
+ - lgy0404/MemGUI-3K
5
  language:
6
+ - en
7
  library_name: transformers
8
+ license: apache-2.0
9
  pipeline_tag: image-text-to-text
 
 
 
10
  tags:
11
+ - qwen3_vl
12
+ - gui-agent
13
+ - mobile-gui
14
+ - android
15
+ - memory
16
+ - context-management
17
+ - conact
18
+ - memgui-agent
19
+ - long-horizon
20
  model-index:
21
+ - name: MemGUI-8B-SFT
22
+ results:
23
+ - task:
24
+ type: image-text-to-text
25
+ name: Long-horizon mobile GUI control
26
+ dataset:
27
+ name: MemGUI-Bench
28
+ type: lgy0404/MemGUI-3K
29
+ metrics:
30
+ - type: pass_at_1
31
+ value: 23.4
32
+ name: Pass@1
33
+ - type: pass_at_3
34
+ value: 35.9
35
+ name: Pass@3
36
+ - type: irr
37
+ value: 30.2
38
+ name: Information Retention Rate
39
+ - task:
40
+ type: image-text-to-text
41
+ name: Out-of-distribution mobile GUI control
42
+ dataset:
43
+ name: MobileWorld GUI-Only
44
+ type: mobileworld-gui-only
45
+ metrics:
46
+ - type: success_rate
47
+ value: 17.9
48
+ name: Success Rate
 
 
 
 
49
  ---
50
+
51
  # MemGUI-8B-SFT
52
 
53
+ [**Project Page**](https://memgui-agent.github.io/) | [**Paper**](https://huggingface.co/papers/2606.19926) | [**Code**](https://github.com/kwai/MemGUI-Agent)
54
+
55
  MemGUI-8B-SFT is an 8B MemGUI-Agent model trained from Qwen3-VL-8B-Instruct on
56
  MemGUI-3K. It is designed for long-horizon mobile GUI control with proactive
57
  context management.
 
64
 
65
  ## Model Details
66
 
67
+ - **Model type:** multimodal mobile GUI agent
68
+ - **Base model:** `Qwen/Qwen3-VL-8B-Instruct`
69
+ - **Training data:** `lgy0404/MemGUI-3K`
70
+ - **Training recipe:** supervised fine-tuning with ms-swift
71
+ - **Output protocol:** ConAct 5-part structured output
72
+ - **License:** Apache 2.0
73
 
74
  ## Intended Use
75
 
 
130
  ```bibtex
131
  @article{memguiagent2026,
132
  title = {MemGUI-Agent: An End-to-End Long-Horizon Mobile GUI Agent with Proactive Context Management},
133
+ author = {Guangyi Liu and Gao Wu and Congxiao Liu and Pengxiang Zhao and Liang Liu and Mading Li and Qi Zhang and Mengyan Wang and Liang Guo and Yong Liu},
134
+ year = {2026},
135
+ journal = {arXiv preprint arXiv:2606.19926}
136
  }
137
+ ```