Datasetpublic

webbrain-one/webbrain-vl-2-450M-dataset

webbrain-vl-2-450M-dataset Browser viewport screenshots paired with the six-section observation format used by WebBrain's vision subsystem. Labels were generated by a teacher VLM with DOM/accessibility evidence, then filtered for structure and grounded exact text. This dataset trains webbrain-one/webbrain-vl-2-450M. The browser-ready export is webbrain-one/webbrain-vl-2-450M-onnx. Schema image: browser viewport screenshot system_prompt: WebBrain production vision… See the full description on the dataset page: https://huggingface.co/datasets/webbrain-one/webbrain-vl-2-450M-dataset.

sourceHugging Faceotherupdated 1mo agoView on Hugging Face

1likes364downloads

Dataset Card

webbrain-vl-2-450M-dataset

Browser viewport screenshots paired with the six-section observation format used by WebBrain's vision subsystem. Labels were generated by a teacher VLM with DOM/accessibility evidence, then filtered for structure and grounded exact text.

This dataset trains `webbrain-one/webbrain-vl-2-450M`. The browser-ready export is `webbrain-one/webbrain-vl-2-450M-onnx`.

Schema

—image: browser viewport screenshot
—system_prompt: WebBrain production vision system prompt
—user_prompt: WebBrain production screenshot request
—response: accepted six-section teacher label
—messages_json: leap-finetune-compatible multimodal conversation
—group_id: split/leakage group
—image_sha256: image identity
—source_*: source, license, URL, and provenance
—teacher_model: teacher identifier; private endpoint details are not included

Splits

—train: 46,879
—validation: 3,121

Splits are grouped by task/site and deduplicated by image SHA-256. WebBrain's 100-case test/vision suite is excluded and remains held out.

Sources and licenses

Source	License	Rows
`Farama-Foundation/miniwob-plusplus`	MIT	4,592
`HuggingFaceM4/WebSight`	CC-BY-4.0	9,561
`docling-project/screenparse`	CC-BY-4.0	2,493
`webbrain-one/targeted-browser-gui-synthetic`	CC-BY-4.0	33,354

This combined dataset uses license: other because source licenses differ. Users must comply with every row's source_license and the linked upstream terms. WebSight use must be disclosed and attributed under CC-BY-4.0. MiniWoB++ is MIT. ScreenParse declares CC-BY-4.0; preserve its attribution and review sampled third-party page content. Project-generated synthetic GUI rows are CC-BY-4.0. Do not assume project-owned or authorized captures grant rights beyond their recorded license.

Teacher

—Qwen/Qwen3.6-35B-A3B-Q4_K_M: 366 accepted labels
—qwen/qwen3.6-35b-a3b: 49,634 accepted labels

Teacher outputs can contain OCR errors or hallucinations. The automated quality gate does not replace manual review. Do not use this dataset for security, payment, identity, or accessibility decisions without independent validation.

Intended use

Supervised fine-tuning and evaluation of compact VLMs that turn browser screenshots into structured observations for a downstream automation planner. It is not a next-action or click-coordinate dataset.

The image column is embedded in Hub Parquet files. To recreate loadable local image paths and typed messages JSONL for LiquidAI/leap-finetune, use this project's scripts/materialize_hf_dataset.py utility.