Image-Text-to-Text
Transformers
Safetensors
English
qwen2_5_vl
agent
conversational
text-generation-inference
Instructions to use OpenGVLab/ScaleCUA-7B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use OpenGVLab/ScaleCUA-7B with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="OpenGVLab/ScaleCUA-7B") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] pipe(text=messages)# Load model directly from transformers import AutoProcessor, AutoModelForImageTextToText processor = AutoProcessor.from_pretrained("OpenGVLab/ScaleCUA-7B") model = AutoModelForImageTextToText.from_pretrained("OpenGVLab/ScaleCUA-7B") messages = [ { "role": "user", "content": [ {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"}, {"type": "text", "text": "What animal is on the candy?"} ] }, ] inputs = processor.apply_chat_template( messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", ).to(model.device) outputs = model.generate(**inputs, max_new_tokens=40) print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:])) - Notebooks
- Google Colab
- Kaggle
- Local Apps
- vLLM
How to use OpenGVLab/ScaleCUA-7B with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "OpenGVLab/ScaleCUA-7B" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "OpenGVLab/ScaleCUA-7B", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker
docker model run hf.co/OpenGVLab/ScaleCUA-7B
- SGLang
How to use OpenGVLab/ScaleCUA-7B with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "OpenGVLab/ScaleCUA-7B" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "OpenGVLab/ScaleCUA-7B", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "OpenGVLab/ScaleCUA-7B" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "OpenGVLab/ScaleCUA-7B", "messages": [ { "role": "user", "content": [ { "type": "text", "text": "Describe this image in one sentence." }, { "type": "image_url", "image_url": { "url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg" } } ] } ] }' - Docker Model Runner
How to use OpenGVLab/ScaleCUA-7B with Docker Model Runner:
docker model run hf.co/OpenGVLab/ScaleCUA-7B
Improve model card: Update paper links and add project page
#1
by nielsr HF Staff - opened
README.md
CHANGED
|
@@ -1,24 +1,25 @@
|
|
| 1 |
---
|
| 2 |
-
|
|
|
|
| 3 |
datasets:
|
| 4 |
- OpenGVLab/ScaleCUA-Data
|
| 5 |
language:
|
| 6 |
- en
|
|
|
|
|
|
|
| 7 |
metrics:
|
| 8 |
- accuracy
|
| 9 |
-
base_model:
|
| 10 |
-
- Qwen/Qwen2.5-VL-7B-Instruct
|
| 11 |
pipeline_tag: image-text-to-text
|
| 12 |
-
|
| 13 |
tags:
|
| 14 |
- agent
|
| 15 |
---
|
| 16 |
|
| 17 |
# SCALECUA: SCALING UP COMPUTER USE AGENTS WITH CROSS-PLATFORM DATA
|
| 18 |
|
| 19 |
-
|
| 20 |
-
|
| 21 |
|
|
|
|
| 22 |
|
| 23 |
## Introduction
|
| 24 |
|
|
@@ -315,7 +316,8 @@ Previous operations:
|
|
| 315 |
def format_history(history):
|
| 316 |
if len(history) > 0:
|
| 317 |
actions_history = [f"Step {i+1}: {low_level}" for i, low_level in enumerate(history)]
|
| 318 |
-
return "
|
|
|
|
| 319 |
else:
|
| 320 |
return None
|
| 321 |
|
|
@@ -382,7 +384,8 @@ def parse_response(response: str) -> Dict:
|
|
| 382 |
if action_matches:
|
| 383 |
for match in action_matches:
|
| 384 |
# Split each match by newline and strip whitespace from each line
|
| 385 |
-
lines = [line.strip() for line in match.split('
|
|
|
|
| 386 |
actions.extend(lines)
|
| 387 |
operation_match = re.search(r'<operation>\s*(.*?)\s*</operation>', response, re.DOTALL)
|
| 388 |
operation = operation_match.group(1).strip() if operation_match else None
|
|
@@ -456,12 +459,12 @@ def parse_actions(self, actions):
|
|
| 456 |
|
| 457 |
else:
|
| 458 |
if "=" in args_str:
|
| 459 |
-
for arg in re.finditer(r"(\w+)=\[([^\]]+)\]", args_str):
|
| 460 |
param = arg.group(1)
|
| 461 |
list_str = arg.group(2)
|
| 462 |
|
| 463 |
list_items = []
|
| 464 |
-
for item in re.finditer(r"'([^']*)'|\"([^\"]*)\"|([^,\]]+)", list_str):
|
| 465 |
val = (item.group(1) or item.group(2) or item.group(3)).strip()
|
| 466 |
if val:
|
| 467 |
list_items.append(val.strip('"\''))
|
|
@@ -469,7 +472,7 @@ def parse_actions(self, actions):
|
|
| 469 |
args[param] = list_items
|
| 470 |
|
| 471 |
|
| 472 |
-
for arg in re.finditer(r"(\w+)=([^,)]+)", args_str):
|
| 473 |
param = arg.group(1)
|
| 474 |
if param in args:
|
| 475 |
continue
|
|
@@ -490,7 +493,7 @@ def parse_actions(self, actions):
|
|
| 490 |
|
| 491 |
else:
|
| 492 |
args_list = []
|
| 493 |
-
for arg in re.finditer(r"'([^']*)'|\"([^\"]*)\"|([^,]+)", args_str):
|
| 494 |
val = (arg.group(1) or arg.group(2) or arg.group(3)).strip()
|
| 495 |
if val:
|
| 496 |
args_list.append(val.strip('"\''))
|
|
@@ -532,6 +535,8 @@ If you find our project useful in your research, please consider citing:
|
|
| 532 |
title = {ScaleCUA: Scaling Open-Source Computer Use Agents with Cross-Platform Data},
|
| 533 |
author = {Liu, Zhaoyang and Xie, Jingjing and Ding, Zichen and Li, Zehao and Yang, Bowen and Wu, Zhenyu and Wang, Xuehui and Sun, Qiushi and Liu, Shi and Wang, Weiyun and Ye, Shenglong and Li, Qingyun and Dong, Xuan and Yu, Yue and Lu, Chenyu and Mo, YunXiang and Yan, Yao and Tian, Zeyue and Zhang, Xiao and Huang, Yuan and Liu, Yiqian and Su, Weijie and Luo, Gen and Yue, Xiangyu and Qi, Biqing and Chen, Kai and Zhou, Bowen and Qiao, Yu and Chen, Qifeng and Wang, Wenhai},
|
| 534 |
year = {2025},
|
| 535 |
-
|
|
|
|
|
|
|
| 536 |
}
|
| 537 |
```
|
|
|
|
| 1 |
---
|
| 2 |
+
base_model:
|
| 3 |
+
- Qwen/Qwen2.5-VL-7B-Instruct
|
| 4 |
datasets:
|
| 5 |
- OpenGVLab/ScaleCUA-Data
|
| 6 |
language:
|
| 7 |
- en
|
| 8 |
+
library_name: transformers
|
| 9 |
+
license: apache-2.0
|
| 10 |
metrics:
|
| 11 |
- accuracy
|
|
|
|
|
|
|
| 12 |
pipeline_tag: image-text-to-text
|
| 13 |
+
project_page: https://osatlas.github.io/
|
| 14 |
tags:
|
| 15 |
- agent
|
| 16 |
---
|
| 17 |
|
| 18 |
# SCALECUA: SCALING UP COMPUTER USE AGENTS WITH CROSS-PLATFORM DATA
|
| 19 |
|
| 20 |
+
This model is the official implementation of the paper [ScaleCUA: Scaling Open-Source Computer Use Agents with Cross-Platform Data](https://huggingface.co/papers/2509.15221).
|
|
|
|
| 21 |
|
| 22 |
+
[\[π GitHub\]](https://github.com/OpenGVLab/ScaleCUA) [\[π Paper\]](https://huggingface.co/papers/2509.15221) [\[π Project Page\]](https://osatlas.github.io/) [\[π Quick Start\]](#model-loading)
|
| 23 |
|
| 24 |
## Introduction
|
| 25 |
|
|
|
|
| 316 |
def format_history(history):
|
| 317 |
if len(history) > 0:
|
| 318 |
actions_history = [f"Step {i+1}: {low_level}" for i, low_level in enumerate(history)]
|
| 319 |
+
return "
|
| 320 |
+
".join(actions_history)
|
| 321 |
else:
|
| 322 |
return None
|
| 323 |
|
|
|
|
| 384 |
if action_matches:
|
| 385 |
for match in action_matches:
|
| 386 |
# Split each match by newline and strip whitespace from each line
|
| 387 |
+
lines = [line.strip() for line in match.split('
|
| 388 |
+
') if line.strip()]
|
| 389 |
actions.extend(lines)
|
| 390 |
operation_match = re.search(r'<operation>\s*(.*?)\s*</operation>', response, re.DOTALL)
|
| 391 |
operation = operation_match.group(1).strip() if operation_match else None
|
|
|
|
| 459 |
|
| 460 |
else:
|
| 461 |
if "=" in args_str:
|
| 462 |
+
for arg in re.finditer(r"(\w+)=\\[([^\\]]+)\\]", args_str):
|
| 463 |
param = arg.group(1)
|
| 464 |
list_str = arg.group(2)
|
| 465 |
|
| 466 |
list_items = []
|
| 467 |
+
for item in re.finditer(r"'([^']*)'|\"([^\\\"]*)\"|([^,\\]]+)", list_str):
|
| 468 |
val = (item.group(1) or item.group(2) or item.group(3)).strip()
|
| 469 |
if val:
|
| 470 |
list_items.append(val.strip('"\''))
|
|
|
|
| 472 |
args[param] = list_items
|
| 473 |
|
| 474 |
|
| 475 |
+
for arg in re.finditer(r"(\\w+)=([^,)]+)", args_str):
|
| 476 |
param = arg.group(1)
|
| 477 |
if param in args:
|
| 478 |
continue
|
|
|
|
| 493 |
|
| 494 |
else:
|
| 495 |
args_list = []
|
| 496 |
+
for arg in re.finditer(r"'([^']*)'|\"([^\\\"]*)\"|([^,]+)", args_str):
|
| 497 |
val = (arg.group(1) or arg.group(2) or arg.group(3)).strip()
|
| 498 |
if val:
|
| 499 |
args_list.append(val.strip('"\''))
|
|
|
|
| 535 |
title = {ScaleCUA: Scaling Open-Source Computer Use Agents with Cross-Platform Data},
|
| 536 |
author = {Liu, Zhaoyang and Xie, Jingjing and Ding, Zichen and Li, Zehao and Yang, Bowen and Wu, Zhenyu and Wang, Xuehui and Sun, Qiushi and Liu, Shi and Wang, Weiyun and Ye, Shenglong and Li, Qingyun and Dong, Xuan and Yu, Yue and Lu, Chenyu and Mo, YunXiang and Yan, Yao and Tian, Zeyue and Zhang, Xiao and Huang, Yuan and Liu, Yiqian and Su, Weijie and Luo, Gen and Yue, Xiangyu and Qi, Biqing and Chen, Kai and Zhou, Bowen and Qiao, Yu and Chen, Qifeng and Wang, Wenhai},
|
| 537 |
year = {2025},
|
| 538 |
+
journal = {arXiv preprint arXiv:2509.15221},
|
| 539 |
+
note = {Preprint},
|
| 540 |
+
url = {https://arxiv.org/abs/2509.15221}
|
| 541 |
}
|
| 542 |
```
|