Jev directory
CommunityJev Models & Reimplementations
Jev GUI Training Preview
Hugging Face dataset from IntelligenceLab holding 100K-1M GUI-agent decision samples for Jev-style models. Each row pairs screenshots with a goal state, a multiple-choice question, an answer index, and metadata such as action types, horizon, and negative-plan recipes.
- Reading time
- 2 min
- Published by
- Community
- Added
- Nov 2026
Highlights
- Hosted on Hugging Face as IntelligenceLab/Jev-GUI-Training-Preview, sized 100K-1M rows.
- Rows pair image IDs with a goal state, a question, a list of options, and a numeric answer index.
- Metadata records action types, domain, horizon, and negative-plan recipes per sample.
- Distributed in parquet and tagged gui-agent, multimodal, decision-model, jev, and qwen.
- Sources include aguvis, miniwob, and guiact-web-single trajectories; strata include gui.teacher and gui.short_horizon.
Questions about this entry
Short answers drawn from the linked project or page.
It is a Hugging Face dataset published by IntelligenceLab under the Jev name, tagged gui-agent, multimodal, decision-model, jev, and qwen. It holds GUI-agent decision samples formatted as multiple-choice questions over screenshots, with a goal state, options, and an answer index.
The dataset card lists parquet as the format and the Datasets, Dask, and Polars libraries, so it can be read with those tools from the Hugging Face Hub. Files and versions are browsable on the dataset page, and the card also exposes a Data Studio view.
Rows include a state_json with the goal and previous actions, a question string, a list of options, a numeric answer index, image_ids for the referenced screenshots, and a meta_json with fields such as action_types, domain, horizon, and negative_recipe.
The dataset card lists the license as other. Because that is not a standard named license, check the card and the publishing organization for the exact terms before redistributing or using the data in a product.
The card lists English as the language and image, tabular, and text as modalities, with the task set to Image-Text-to-Text. Samples reference one or two screenshots per decision, so image input is part of the record rather than an optional extra.
The card reports a size band of 100K to 1M rows. Sample indexes visible in the preview run into the low double digits, and each row carries its own sample_index, id, group, source, stratum, and family fields.
It is training and evaluation data, not a runtime, SDK, or inference endpoint, so it cannot answer decisions on its own. It also targets GUI action selection rather than chat or free-text generation, and the preview label suggests it is not a final frozen release.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
Hugging Face dataset of Jev-shaped retrieval and classification decisions built from MM-BRIGHT, holding Qwen3.6-35B-A3B-FP8 state, action, and hard-negative embeddings for post-training CLM heads.
Jev Models & Reimplementations#open-models#benchmark#evaluation
Community
Hugging Face dataset that renders HoVer claim-verification items as Jev Decision Index retrieval and classification requests, with Qwen3.6-35B-A3B-FP8 state, action, and hard-negative embeddings for post-training CLM heads.
Jev Models & Reimplementations#classification#benchmark#open-models
Community
arXiv paper testing whether the readout and adaptation choices used for direct selection still hold up in cross-request entailment reranking for JEV decision models. It combines a mean-span readout with selective backbone adaptation and reports accuracy on MC11, an average over eleven reranking configurations.
Jev Models & ReimplementationsArticle#arxiv#research#classification
Community
Embedding dataset from clm-project that renders NVIDIA's When2Call training split into Jev Decision Index-style tool-use decisions and stores Qwen3.6-35B-A3B-FP8 state, action, and hard-negative vectors for post-training CLM heads.
Jev Models & Reimplementations#open-models#research#agents
Community
A family of 0.8B-9B ophthalmic decision models that score candidate answers without autoregressive decoding, trained on RetiJev-Data: 78,158 clinical-decision and diagnosis questions and 50,074 guideline-grading questions. RetiJev-9B answers a case in 158 ms on one GPU.
Jev Models & ReimplementationsArticle#arxiv#open-models#benchmark
Community
Hugging Face dataset of Qwen3.6-35B-A3B-FP8 state/action embeddings for contrastive post-training on the Jev Decision Index's entailment and verification train splits: ANLI, ContractNLI, NLI4CT, HoVer and RAGTruth.
Jev Models & Reimplementations#open-models#calibration#research
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.