Jev directory
CommunityJev Models & Reimplementations
Kieve-JeV MoE 300M
An encoder-based typed decision engine: given a state and typed questions, it returns one answer per question with a calibrated probability distribution over the supplied options. Built on a frozen ModernBERT-base backbone with rank-16 LoRA adapters and a separate noul head.
- Reading time
- 2 min
- Published by
- Community
- Added
- Oct 2026
Highlights
- Total parameters: 293.9M, with a frozen ModernBERT-base backbone and rank-16 LoRA adapters.
- Returns calibrated distributions only over the options supplied and never invents options.
- Variant includes state and question encoders, a paired cross-attention option scorer, and a separate noul head.
- Held-out JvD evaluation reports atbench500 0.610 and jailbreak 0.638.
- Decision Index 0.2.1 reports decision_index 1.05 and raw 22.41 across 150,759 requests.
Questions about this entry
Short answers drawn from the linked project or page.
It is an encoder-based typed decision engine. It takes a state, which can be any text, plus a set of typed questions, and returns for every question one answer with a calibrated probability distribution over the supplied options, or a probability for the yes/no noul primitive.
No. It selects among the options you pass in and never invents options, so the distribution is defined strictly over those options. Tasks that require generating open-ended text are outside what this decision engine returns.
The backbone is a frozen answerdotai/ModernBERT-base with LoRA adapters of rank 16. The jev variant adds state and question encoders, a paired cross-attention option scorer, and a separate noul head, using fused neural operators plus the stack described in its config.
The repository provides checkpoint_latest.pt, a training checkpoint whose model_state holds weights and a temperature calibrated for softmax, plus config.json, the inference config consumed by Kieve.
The model is released under Apache-2.0. The ModernBERT backbone it builds on is subject to its own separate license, so both apply when you use the checkpoint.
Held-out JvD run metrics report atbench500 at 0.610 and jailbreak at 0.638. The model card also lists Decision Index 0.2.1 with decision_index 1.05, raw 22.41, 150,759 requests, marked complete.
No. The model page states that it is not deployed by any Inference Provider, though a request for provider support can be raised from the page; otherwise you load the checkpoint yourself.
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 26M-class decision model on Hugging Face, fine-tuned from q-project/N-Q-26M-Chat. It ships a custom NQ4ForDecision class for text classification and feature extraction, is tagged calibrated and decision-model, and publishes Apache-2.0 weights with Russian and English support.
Jev Models & Reimplementations#open-models#calibration#classification
Community
Hugging Face dataset of Qwen3.6-35B-A3B-FP8 state/action embeddings for post-training contrastive heads on the Jev Decision Index's Knowledge & Reasoning train splits: CommonsenseQA, MMLU auxiliary_train and ChessBench.
Jev Models & Reimplementations#open-models#calibration#research
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
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
Russian-first open typed-decision model on Hugging Face. Given a context and a set of candidate options it returns a calibrated probability and picks one, covering yes/no, choice, and ordinal scale decisions. Built on mmBERT-base with 307M parameters under an MIT license.
Jev Models & Reimplementations#open-models#calibration#classification
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
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.