Datasets:
metadata
dataset_info:
features:
- name: task_id
dtype: string
- name: platform
dtype:
class_label:
names:
'0': 7-Zip
'1': Affine
'2': Atom
'3': Anki
'4': Bash
'5': Blender
'6': Chromium
'7': Element
'8': Eclipse
'9': FreeCAD
'10': Gedit
'11': GIMP
'12': Inkscape
'13': IntelliJ IDEA
'14': LibreOffice Calc
'15': LibreOffice Draw
'16': LibreOffice Impress
'17': LibreOffice Writer
'18': Mastodon
'19': Matrix
'20': Nextcloud
'21': OnlyOffice Calendar
'22': OnlyOffice Presentation
'23': OBS Studio
- name: instruction
dtype: string
- name: step_index
dtype: int32
- name: image
dtype: image
- name: code
dtype: string
- name: action
dtype: string
splits:
- name: train
num_bytes: 2854516158
num_examples: 8599
- name: test
num_bytes: 864575732
num_examples: 2355
download_size: 4099027918
dataset_size: 3719091890
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
tags:
- computer-use-agent
Trace2Action Dataset
Trace2Action is a large-scale computer-use dataset designed for training and evaluating multimodal agents that operate graphical user interfaces through grounded action generation and procedural reasoning.
Structure
| Field | Type | Description |
|---|---|---|
task_id |
string | Unique trajectory identifier |
platform |
string | Application platform name |
instruction |
string | Natural language task objective |
step_index |
int32 | Temporal index within trajectory |
image |
Image | Screenshot corresponding to current state |
code |
string | Ground-truth executable UI action |
action |
json | Action description in JSON format |
Action space
{"type": "CLICK", "x": [0, 100], "y": [0, 100] }{"type": "RIGHT_CLICK", "x": [0, 100], "y": [0, 100]}{"type": "DOUBLE_CLICK", "x": [0, 100], "y": [0, 100]}{"type": "DRAG_TO", "x": [0, 100], "y": [0, 100]}{"type": "SCROLL", "amount": <int>}{"type": "PRESS", "key": "<str>"}{"type": "HOTKEY", "keys": ["<str>", "<str>"]}{"type": "TYPE", "text": "<str>"}
Applications
Train set (20):
- 7-Zip
- Affine
- Atom
- Anki
- Bash
- Blender
- Chromium
- Element
- Eclipse
- FreeCAD
- GIMP
- Inkscape
- IntelliJ IDEA
- LibreOffice Calc
- LibreOffice Draw
- LibreOffice Writer
- Mastodon
- Nextcloud
- OnlyOffice Calendar
- OnlyOffice Presentation
Test set (4):
- Gedit
- LibreOffice Impress
- Matrix
- OBS Studio
Download
from datasets import load_dataset
ds = load_dataset("dyfang/trace2action-10k")