Datasets:
Add dataset card, provenance, and release metadata
Browse files- LICENSES.md +20 -0
- README.md +96 -0
- manifest_meta.json +60 -0
- release_stats.json +30 -0
- training_mix_gui_v1.json +70 -0
LICENSES.md
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Source terms and redistribution warning
|
| 2 |
+
|
| 3 |
+
This collection has **mixed and unresolved source terms**. The repository uses
|
| 4 |
+
`license: other`; that is a warning, not a new license grant.
|
| 5 |
+
|
| 6 |
+
- [xlangai/aguvis-stage2](https://huggingface.co/datasets/xlangai/aguvis-stage2)
|
| 7 |
+
did not declare a dataset license on the source card at the pinned revision.
|
| 8 |
+
- [agentsea/wave-ui-25k](https://huggingface.co/datasets/agentsea/wave-ui-25k)
|
| 9 |
+
did not declare a dataset license on the source card at the pinned revision.
|
| 10 |
+
- [tasksource/tasksource-jev-typed-decisions](https://huggingface.co/datasets/tasksource/tasksource-jev-typed-decisions)
|
| 11 |
+
is a mixed-license collection and explicitly warns that visual annotation
|
| 12 |
+
terms may not cover third-party images and that some image rights remain
|
| 13 |
+
unresolved.
|
| 14 |
+
- Teacher-generated questions, options and probability targets do not replace
|
| 15 |
+
or broaden the rights attached to source screenshots.
|
| 16 |
+
|
| 17 |
+
GUI screenshots and traces may contain third-party copyrighted interfaces,
|
| 18 |
+
usernames, email addresses or other personal-looking text. Users are
|
| 19 |
+
responsible for reviewing applicable source terms, privacy requirements and
|
| 20 |
+
their intended use.
|
README.md
ADDED
|
@@ -0,0 +1,96 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: Jev GUI Training Preview
|
| 3 |
+
license: other
|
| 4 |
+
language:
|
| 5 |
+
- en
|
| 6 |
+
task_categories:
|
| 7 |
+
- image-text-to-text
|
| 8 |
+
tags:
|
| 9 |
+
- gui-agent
|
| 10 |
+
- multimodal
|
| 11 |
+
- decision-model
|
| 12 |
+
- jev
|
| 13 |
+
- qwen
|
| 14 |
+
size_categories:
|
| 15 |
+
- 100K<n<1M
|
| 16 |
+
configs:
|
| 17 |
+
- config_name: decisions
|
| 18 |
+
data_files:
|
| 19 |
+
- split: train
|
| 20 |
+
path: decisions/train-*.parquet
|
| 21 |
+
- config_name: images
|
| 22 |
+
data_files:
|
| 23 |
+
- split: train
|
| 24 |
+
path: images/train-*.parquet
|
| 25 |
+
---
|
| 26 |
+
|
| 27 |
+
# Jev GUI Training Preview
|
| 28 |
+
|
| 29 |
+
This public release contains the GUI-only weighted training mixture prepared
|
| 30 |
+
for Jev-Vision:
|
| 31 |
+
|
| 32 |
+
- **636,800 decision references**
|
| 33 |
+
- **132,647 unique images**, stored once
|
| 34 |
+
- Choice, yes/no (`noul`) and ordered-score decisions
|
| 35 |
+
- Single-step, short-horizon, teacher-generated and UI-element tasks
|
| 36 |
+
|
| 37 |
+
The weighted sampler intentionally repeats some records. `sample_index` is
|
| 38 |
+
unique, while `id` identifies the underlying decision and may repeat.
|
| 39 |
+
|
| 40 |
+
## Configurations
|
| 41 |
+
|
| 42 |
+
### `decisions`
|
| 43 |
+
|
| 44 |
+
One row per sampled training reference. `image_ids` contains ordered content
|
| 45 |
+
hashes referring to rows in the `images` configuration. `state_json` and
|
| 46 |
+
`meta_json` preserve structured fields as JSON strings for a stable Parquet
|
| 47 |
+
schema.
|
| 48 |
+
|
| 49 |
+
### `images`
|
| 50 |
+
|
| 51 |
+
One row per unique image. The `id` column is the content hash used by
|
| 52 |
+
`decisions.image_ids`; `image` is a Hugging Face Image feature.
|
| 53 |
+
|
| 54 |
+
```python
|
| 55 |
+
from datasets import load_dataset
|
| 56 |
+
|
| 57 |
+
decisions = load_dataset(
|
| 58 |
+
"IntelligenceLab/Jev-GUI-Training-Preview",
|
| 59 |
+
"decisions",
|
| 60 |
+
split="train",
|
| 61 |
+
)
|
| 62 |
+
images = load_dataset(
|
| 63 |
+
"IntelligenceLab/Jev-GUI-Training-Preview",
|
| 64 |
+
"images",
|
| 65 |
+
split="train",
|
| 66 |
+
)
|
| 67 |
+
```
|
| 68 |
+
|
| 69 |
+
## Sample composition
|
| 70 |
+
|
| 71 |
+
| Source | Sampled rows |
|
| 72 |
+
|---|---:|
|
| 73 |
+
| `aguvis` | 180,087 |
|
| 74 |
+
| `aguvis_short_horizon` | 213,646 |
|
| 75 |
+
| `tasksource_vision` | 4,967 |
|
| 76 |
+
| `teacher_gui` | 126,723 |
|
| 77 |
+
| `teacher_stageb` | 51,390 |
|
| 78 |
+
| `waveui` | 59,987 |
|
| 79 |
+
|
| 80 |
+
The exact weighted mixture is in `training_mix_gui_v1.json`; build metadata is
|
| 81 |
+
in `manifest_meta.json` and `release_stats.json`.
|
| 82 |
+
|
| 83 |
+
## Important rights and privacy notice
|
| 84 |
+
|
| 85 |
+
This release has mixed and unresolved source-image rights and may include
|
| 86 |
+
personal-looking text present in public GUI traces. Read `LICENSES.md` before
|
| 87 |
+
use. `license: other` does not grant rights beyond each upstream source.
|
| 88 |
+
|
| 89 |
+
## Provenance
|
| 90 |
+
|
| 91 |
+
- [AGUVIS Stage 2](https://huggingface.co/datasets/xlangai/aguvis-stage2)
|
| 92 |
+
- [Wave-UI 25K](https://huggingface.co/datasets/agentsea/wave-ui-25k)
|
| 93 |
+
- [Tasksource Jev Typed Decisions](https://huggingface.co/datasets/tasksource/tasksource-jev-typed-decisions)
|
| 94 |
+
- Teacher-generated GUI decisions from Qwen3.5-122B-A10B
|
| 95 |
+
- Deterministic seed: `20261027`
|
| 96 |
+
- Mix configuration SHA-256: `632255ca91150dd16e46f0717bb2433866d38bc5d9a48656ba1aa3d07c6b483a`
|
manifest_meta.json
ADDED
|
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"version": "visual-jev-gui-original-v1",
|
| 3 |
+
"seed": 20261027,
|
| 4 |
+
"total_samples": 636800,
|
| 5 |
+
"world_size": 64,
|
| 6 |
+
"samples_per_rank": 9950,
|
| 7 |
+
"files": [
|
| 8 |
+
"records/aguvis/train.jsonl",
|
| 9 |
+
"records/aguvis_short_horizon/train.jsonl",
|
| 10 |
+
"records/tasksource_vision/train.jsonl",
|
| 11 |
+
"records/teacher_gui/train.jsonl",
|
| 12 |
+
"records/teacher_stageb/train.jsonl",
|
| 13 |
+
"records/waveui/train.jsonl"
|
| 14 |
+
],
|
| 15 |
+
"source_inventory": {
|
| 16 |
+
"aguvis": 180115,
|
| 17 |
+
"aguvis_short_horizon": 213636,
|
| 18 |
+
"tasksource_vision": 4991,
|
| 19 |
+
"teacher_gui": 126716,
|
| 20 |
+
"teacher_stageb": 51379,
|
| 21 |
+
"waveui": 59975
|
| 22 |
+
},
|
| 23 |
+
"buckets": {
|
| 24 |
+
"gui.aguvis_step|choice|original": {
|
| 25 |
+
"available_records": 180115,
|
| 26 |
+
"sampled_records": 180087
|
| 27 |
+
},
|
| 28 |
+
"gui.rico|choice|original": {
|
| 29 |
+
"available_records": 4991,
|
| 30 |
+
"sampled_records": 4967
|
| 31 |
+
},
|
| 32 |
+
"gui.short_horizon|choice|original": {
|
| 33 |
+
"available_records": 213636,
|
| 34 |
+
"sampled_records": 213646
|
| 35 |
+
},
|
| 36 |
+
"gui.stageb|choice|original": {
|
| 37 |
+
"available_records": 16535,
|
| 38 |
+
"sampled_records": 30834
|
| 39 |
+
},
|
| 40 |
+
"gui.stageb|noul|original": {
|
| 41 |
+
"available_records": 16289,
|
| 42 |
+
"sampled_records": 15417
|
| 43 |
+
},
|
| 44 |
+
"gui.stageb|score|original": {
|
| 45 |
+
"available_records": 18555,
|
| 46 |
+
"sampled_records": 5139
|
| 47 |
+
},
|
| 48 |
+
"gui.teacher|choice|original": {
|
| 49 |
+
"available_records": 126716,
|
| 50 |
+
"sampled_records": 126723
|
| 51 |
+
},
|
| 52 |
+
"gui.waveui|choice|original": {
|
| 53 |
+
"available_records": 59975,
|
| 54 |
+
"sampled_records": 59987
|
| 55 |
+
}
|
| 56 |
+
},
|
| 57 |
+
"config_sha256": "632255ca91150dd16e46f0717bb2433866d38bc5d9a48656ba1aa3d07c6b483a",
|
| 58 |
+
"created_at": "2026-10-28T00:25:43",
|
| 59 |
+
"minutes": 0.4
|
| 60 |
+
}
|
release_stats.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"rows": 636800,
|
| 3 |
+
"samples_per_rank": 9950,
|
| 4 |
+
"unique_images": 132647,
|
| 5 |
+
"raw_image_bytes": 29709290126,
|
| 6 |
+
"decision_parquet_bytes": 332138666,
|
| 7 |
+
"image_parquet_bytes": 29722443306,
|
| 8 |
+
"image_shards": 172,
|
| 9 |
+
"sources": {
|
| 10 |
+
"aguvis": 180087,
|
| 11 |
+
"aguvis_short_horizon": 213646,
|
| 12 |
+
"tasksource_vision": 4967,
|
| 13 |
+
"teacher_gui": 126723,
|
| 14 |
+
"teacher_stageb": 51390,
|
| 15 |
+
"waveui": 59987
|
| 16 |
+
},
|
| 17 |
+
"decision_types": {
|
| 18 |
+
"choice": 616244,
|
| 19 |
+
"noul": 15417,
|
| 20 |
+
"score": 5139
|
| 21 |
+
},
|
| 22 |
+
"source_revisions": {
|
| 23 |
+
"xlangai/aguvis-stage2": "bd6f1ed43498beb8e5474adcb2592c7d521d692e",
|
| 24 |
+
"agentsea/wave-ui-25k": "1b77b749cb4c9936f35838ed8cb0f1b69957ad08",
|
| 25 |
+
"tasksource/tasksource-jev-typed-decisions": "975339b348eb083931e20cf25e19cefe495024e2"
|
| 26 |
+
},
|
| 27 |
+
"mix_config_sha256": "632255ca91150dd16e46f0717bb2433866d38bc5d9a48656ba1aa3d07c6b483a",
|
| 28 |
+
"built_at": "2026-11-05T07:25:25",
|
| 29 |
+
"total_parquet_bytes": 30054581972
|
| 30 |
+
}
|
training_mix_gui_v1.json
ADDED
|
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"version": "visual-jev-gui-original-v1",
|
| 3 |
+
"seed": 20261027,
|
| 4 |
+
"total_samples": 636800,
|
| 5 |
+
"allow_unmatched": true,
|
| 6 |
+
"type_weights": {
|
| 7 |
+
"choice": 0.60,
|
| 8 |
+
"noul": 0.30,
|
| 9 |
+
"score": 0.10
|
| 10 |
+
},
|
| 11 |
+
"form_weights": {
|
| 12 |
+
"original": 1.0
|
| 13 |
+
},
|
| 14 |
+
"strata": [
|
| 15 |
+
{
|
| 16 |
+
"name": "gui.short_horizon",
|
| 17 |
+
"weight": 0.3355,
|
| 18 |
+
"match": {
|
| 19 |
+
"source": "aguvis_short_horizon",
|
| 20 |
+
"family": "gui",
|
| 21 |
+
"form": "original"
|
| 22 |
+
}
|
| 23 |
+
},
|
| 24 |
+
{
|
| 25 |
+
"name": "gui.aguvis_step",
|
| 26 |
+
"weight": 0.2828,
|
| 27 |
+
"match": {
|
| 28 |
+
"source": "aguvis",
|
| 29 |
+
"family": "gui",
|
| 30 |
+
"form": "original"
|
| 31 |
+
}
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"name": "gui.teacher",
|
| 35 |
+
"weight": 0.1990,
|
| 36 |
+
"match": {
|
| 37 |
+
"source": "teacher_gui",
|
| 38 |
+
"family": "gui",
|
| 39 |
+
"form": "original"
|
| 40 |
+
}
|
| 41 |
+
},
|
| 42 |
+
{
|
| 43 |
+
"name": "gui.waveui",
|
| 44 |
+
"weight": 0.0942,
|
| 45 |
+
"match": {
|
| 46 |
+
"source": "waveui",
|
| 47 |
+
"family": "gui",
|
| 48 |
+
"form": "original"
|
| 49 |
+
}
|
| 50 |
+
},
|
| 51 |
+
{
|
| 52 |
+
"name": "gui.stageb",
|
| 53 |
+
"weight": 0.0807,
|
| 54 |
+
"match": {
|
| 55 |
+
"source": "teacher_stageb",
|
| 56 |
+
"family": "gui",
|
| 57 |
+
"form": "original"
|
| 58 |
+
}
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"name": "gui.rico",
|
| 62 |
+
"weight": 0.0078,
|
| 63 |
+
"match": {
|
| 64 |
+
"source": "tasksource_vision",
|
| 65 |
+
"family": "gui",
|
| 66 |
+
"form": "original"
|
| 67 |
+
}
|
| 68 |
+
}
|
| 69 |
+
]
|
| 70 |
+
}
|