Jev directory
CommunityJev Models & Reimplementations
syvai/jev-dataset
Hugging Face dataset of 1,245,391 synthetic decision-modelling examples carrying Jev teacher probability distributions, used for Claes training. Access is gated and requires manual approval of the repository conditions.
- Reading time
- 2 min
- Published by
- Community
- Added
- Nov 2026
Highlights
- 1,245,391 training rows: 988,000 v1 examples plus 257,391 v2 examples.
- 12,000-row validation split taken from the original frozen, language-stratified v1 holdout.
- Questions generated with gpt-6-luna and answered by jev-1.13.0.
- Targets are soft Jev teacher probability distributions, not human-labelled ground truth.
- Exported in the OneJev unified JSONL training format; total file size 1.32 GB.
Questions about this entry
Short answers drawn from the linked project or page.
The repository is publicly accessible but gated: you must log in or sign up and accept the conditions, and access requests require manual approval. After approval, authenticate with hf auth login and download using hf download syvai/jev-dataset --type dataset --local-dir jev-dataset.
1,245,391 rows: 988,000 v1 examples and 257,391 v2 examples, deterministically shuffled with seed 0. v2 IDs are prefixed with v2: to distinguish the source datasets, and training and validation IDs have zero overlap.
Soft Jev teacher probability distributions, not human-labelled ground truth. Binary targets are the teacher's positive probability and its complement, while choice and score distributions are normalised during conversion. The export retains the distributions but not the original answer confidence or full sampling metadata.
Three: noul for binary decisions with yes and no probability targets, choice for categorical decisions where criteria map option IDs to descriptions, and score for ordinal decisions where criteria are an ordered list of levels and target keys are zero-based level indices encoded as strings.
Questions were generated using gpt-6-luna and answered by jev-1.13.0. Assignments draw from a task pool, a language selection, and taxonomy paths, and the combined data is exported in the OneJev unified JSONL training format.
No. The 12,000-row validation split preserves the original frozen, language-stratified v1 holdout rather than resampling training data, so it does not provide a separate held-out v2 evaluation set.
Agreement with Jev, because the data contains synthetic questions and Jev-generated labels. Language and task coverage reflect the generation policy rather than natural usage frequencies, so results should be read as teacher agreement rather than human-accuracy.
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
A LoRA adapter for Qwen3-Omni-30B-A3B-Instruct that scores a therapist's spoken response on the eight Facilitative Interpersonal Skills items, returning a 1-5 distribution and expected value per item from video, audio, and transcript in one forward pass.
Jev Models & Reimplementations#huggingface#calibration#evaluation
Community
Zuse is a family of three decision models (4B, 9B, 27B) that train a low-rank adapter and pointer head on frozen open-weight bases for 27 European languages. The 27B model reaches 87.5% accuracy with pooled expected calibration error of 0.016 on an 11-source European suite.
Jev Models & ReimplementationsArticle#arxiv#open-models#calibration
Community
Hugging Face dataset of synthetic conversational question-and-answer pairs about Jev internals, covering pointer-head scoring, replay-source counts, slot-selection ladders, and model lineage. English text, 10K-100K rows, Apache-2.0, tagged for fine-tuning and knowledge injection.
Jev Models & Reimplementations#open-models#research#evaluation
Community
Apache-2.0 Hugging Face repository for RSI-JEV-27b, a 27B multimodal decision model based on Qwen3.8-27B that answers supplied multiple-choice questions over images, returning a selected option and probabilities for all options in one forward pass.
Jev Models & Reimplementations#open-models#benchmark#calibration
Community
Open-weight vision-language judge that takes a task instruction plus 10 frames of a robot manipulation attempt and predicts success or failure with calibrated probabilities in a single forward pass. Released with inference code and a 20,628-prediction evaluation set.
Jev Models & ReimplementationsArticle#arxiv#open-models#evaluation
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.