ver Sep2ndv2
Browse filesupdated README with contents about the open world set
- Changelog +6 -0
- README.md +86 -4
- open-world-set.tar.xz +3 -0
Changelog
CHANGED
|
@@ -1,3 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
2024-12-08 Danyang Zhang <[email protected]>
|
| 2 |
|
| 3 |
v1.5
|
|
|
|
| 1 |
+
2025-09-02 Danyang Zhang <[email protected]>
|
| 2 |
+
|
| 3 |
+
v1.6
|
| 4 |
+
|
| 5 |
+
Released the Open World Set
|
| 6 |
+
|
| 7 |
2024-12-08 Danyang Zhang <[email protected]>
|
| 8 |
|
| 9 |
v1.5
|
README.md
CHANGED
|
@@ -8,8 +8,8 @@ license: apache-2.0
|
|
| 8 |
|
| 9 |
WikiHow task set is an InfoUI interaction task set based on
|
| 10 |
[Mobile-Env](https://github.com/X-LANCE/Mobile-Env) proposed in [*Mobile-Env:
|
| 11 |
-
|
| 12 |
-
|
| 13 |
[WikiHow](https://www.wikihow.com/Main-Page) is a collaborative wiki site about
|
| 14 |
various real-life tips with more than 340,000 online articles. To construct the
|
| 15 |
task set, 107,448 pages are crawled, and the dumped website data occupy about
|
|
@@ -159,6 +159,81 @@ canonical set are list in the following table:
|
|
| 159 |
| 11 | `home-search-ingredients` | 5 |
|
| 160 |
| 12 | `home-search-summary` | 10 |
|
| 161 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 162 |
### About
|
| 163 |
|
| 164 |
This task set is developed and maintained by [SJTU
|
|
@@ -170,10 +245,16 @@ using the following BibTeX:
|
|
| 170 |
|
| 171 |
```bibtex
|
| 172 |
@article{DanyangZhang2023_MobileEnv_WikiHow,
|
| 173 |
-
title = {{Mobile-Env}:
|
| 174 |
author = {Danyang Zhang and
|
| 175 |
-
|
|
|
|
|
|
|
|
|
|
| 176 |
Zihan Zhao and
|
|
|
|
|
|
|
|
|
|
| 177 |
Ruisheng Cao and
|
| 178 |
Kai Yu},
|
| 179 |
journal = {CoRR},
|
|
@@ -184,3 +265,4 @@ using the following BibTeX:
|
|
| 184 |
eprint = {2305.08144},
|
| 185 |
}
|
| 186 |
```
|
|
|
|
|
|
| 8 |
|
| 9 |
WikiHow task set is an InfoUI interaction task set based on
|
| 10 |
[Mobile-Env](https://github.com/X-LANCE/Mobile-Env) proposed in [*Mobile-Env:
|
| 11 |
+
Building Qualified Evaluation Benchmarks for LLM-GUI
|
| 12 |
+
Interaction*](https://arxiv.org/abs/2305.08144).
|
| 13 |
[WikiHow](https://www.wikihow.com/Main-Page) is a collaborative wiki site about
|
| 14 |
various real-life tips with more than 340,000 online articles. To construct the
|
| 15 |
task set, 107,448 pages are crawled, and the dumped website data occupy about
|
|
|
|
| 159 |
| 11 | `home-search-ingredients` | 5 |
|
| 160 |
| 12 | `home-search-summary` | 10 |
|
| 161 |
|
| 162 |
+
### The Open World Tasks
|
| 163 |
+
|
| 164 |
+
As supplementary to WikiHow task set, we define 74 open world tasks on a wider
|
| 165 |
+
range of apps. The task definitions are stored under `open-world-set/`. The
|
| 166 |
+
online data for these tasks are not fixed, thus, these tasks will not be
|
| 167 |
+
expected to be reproducible to serve as qualified benchmarks.
|
| 168 |
+
|
| 169 |
+
#### Download the Open World Set
|
| 170 |
+
|
| 171 |
+
To test with the open world set, clone this repo or download
|
| 172 |
+
`open-world-set.tar.xz` and unpack it. Then download
|
| 173 |
+
`open-world-set-avd.tar.xz` and unpack it into a directory, *e.g.*,
|
| 174 |
+
`android-avds`. This Android AVD contains the necessary apps needed by the open
|
| 175 |
+
world set.
|
| 176 |
+
|
| 177 |
+
#### Set Up the Runtime Environment
|
| 178 |
+
|
| 179 |
+
The offered AVD is supposed to work in [Mobile-Env Docker
|
| 180 |
+
Image](https://hub.docker.com/r/zdy023/mobile-env). To use this image, you need
|
| 181 |
+
first clone Mobile-Env repository (>=v4.2). Then launch the container with the command:
|
| 182 |
+
|
| 183 |
+
```sh
|
| 184 |
+
docker run -it --device /dev/kvm --gpus all -v $PWD/android-avds:/root/.android -v $PWD/Mobile-Env:/root/mobile-env -v $PWD/open-world-set:/root/open-world-set zdy023/mobile-env:e928cf1-ubuntu20.04-cu121-android30-torch241
|
| 185 |
+
```
|
| 186 |
+
|
| 187 |
+
Run in the container:
|
| 188 |
+
|
| 189 |
+
```sh
|
| 190 |
+
cd mobile-env && pip install -e . # the dependences have been already installed, this command is to compile the proto files locally in the docker environment
|
| 191 |
+
```
|
| 192 |
+
|
| 193 |
+
Then you can run experiments within this docker image.
|
| 194 |
+
|
| 195 |
+
#### Additional Setup to AVD before Launching Experiments
|
| 196 |
+
|
| 197 |
+
Before experiments, you may need to sign in in Ele.me app in the AVD. To launch
|
| 198 |
+
the AVD in docker with a GUI interface using the host display, we recommend to
|
| 199 |
+
use [Distrobox](https://distrobox.it/).
|
| 200 |
+
|
| 201 |
+
```sh
|
| 202 |
+
distrobox create --root --hostname mobile-env --volume $PWD/android-avds:/root/.android --volume $PWD/Mobile-Env:/root/mobile-env -n mobile-env -i zdy023/mobile-env:e928cf1-ubuntu20.04-cu121-android30-torch241
|
| 203 |
+
distrobox enter --root mobile-env
|
| 204 |
+
|
| 205 |
+
# in the container
|
| 206 |
+
sudo -i --preserve-env=ANDROID_HOME,ADB_MDNS
|
| 207 |
+
|
| 208 |
+
# in `sudo -i`
|
| 209 |
+
export PATH=$ANDROID_HOME/platform-tools:$ANDROID_HOME/emulator:$ANDROID_HOME/cmdline-tools/bin:$PATH
|
| 210 |
+
emulator @Pixel_2_ga_30_op -verbose -no-snapshot
|
| 211 |
+
```
|
| 212 |
+
|
| 213 |
+
#### Other Notices
|
| 214 |
+
|
| 215 |
+
1. Mobile-Env environment should be launched with `restart_simulator_at_reset`
|
| 216 |
+
option, *e.g.*:
|
| 217 |
+
|
| 218 |
+
```python
|
| 219 |
+
env = android_env.load( "/root/open-world-set", "Pixel_2_ga_30_op"
|
| 220 |
+
, run_headless=True, text_model=EasyOCRWrapper(lang_list=["en", "ch_sim"])
|
| 221 |
+
, coordinator_args={"restart_simulator_at_reset": True}
|
| 222 |
+
)
|
| 223 |
+
```
|
| 224 |
+
|
| 225 |
+
2. "Today Weather" app may require proxy to work normally in mainland China. To
|
| 226 |
+
pass the proxy information to Mobile-Env, use `mitm_config` parameter,
|
| 227 |
+
*e.g.*:
|
| 228 |
+
|
| 229 |
+
```python
|
| 230 |
+
env = android_env.load( "/root/open-world-set", "Pixel_2_ga_30_op"
|
| 231 |
+
, run_headless=True, text_model=EasyOCRWrapper(lang_list=["en", "ch_sim"])
|
| 232 |
+
, mitm_config={"method": "syscert", "address": "a.b.c.d", "port": "8000"}
|
| 233 |
+
, coordinator_args={"restart_simulator_at_reset": True}
|
| 234 |
+
)
|
| 235 |
+
```
|
| 236 |
+
|
| 237 |
### About
|
| 238 |
|
| 239 |
This task set is developed and maintained by [SJTU
|
|
|
|
| 245 |
|
| 246 |
```bibtex
|
| 247 |
@article{DanyangZhang2023_MobileEnv_WikiHow,
|
| 248 |
+
title = {{Mobile-Env}: Building Qualified Evaluation Benchmarks for LLM-GUI Interaction},
|
| 249 |
author = {Danyang Zhang and
|
| 250 |
+
Zhennan Shen and
|
| 251 |
+
Rui Xie and
|
| 252 |
+
Situo Zhang and
|
| 253 |
+
Tianbao Xie and
|
| 254 |
Zihan Zhao and
|
| 255 |
+
Siyuan Chen and
|
| 256 |
+
Lu Chen and
|
| 257 |
+
Hongshen Xu and
|
| 258 |
Ruisheng Cao and
|
| 259 |
Kai Yu},
|
| 260 |
journal = {CoRR},
|
|
|
|
| 265 |
eprint = {2305.08144},
|
| 266 |
}
|
| 267 |
```
|
| 268 |
+
|
open-world-set.tar.xz
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:17dd14734e2e49661d71d8f63995b32810917e2ca26b11f94ec6051c6905eaaa
|
| 3 |
+
size 118563144
|