Jev directory
CommunityJev Models & Reimplementations
Femto Jev
A 7.7M-parameter decision model on Hugging Face that reads text or speech and returns one probability per option for choice, score, and noul questions. Speech runs through femto-asr, giving a 10.1M-parameter spoken pipeline.
- Reading time
- 2 min
- Published by
- Community
- Added
- Nov 2026
Highlights
- 7.7M parameters, about 30 MB in fp32; the spoken pipeline with femto-asr totals 10.1M parameters.
- Answers three question types: choice, score, and noul, with questions written as plain text at call time.
- Scores 0.732 accuracy on the typed-decisions official test split of 400 cases and 2,000 decisions.
- Speech can be read three ways from the audio alone: transcript plus frames, transcript only, or frames only.
- On SLURP the frames add 13 points over the transcript; on MInDS-14 they add nothing.
Questions about this entry
Short answers drawn from the linked project or page.
Three types, matching the typed-decisions shape: choice, where each option gets one probability; score, where ordered rubric levels each get one probability; and noul, a yes/no statement returning P(false) and P(true). Questions are plain text written at call time, so nothing is retrained when they change.
The model has 7.7M parameters, roughly 30 MB in fp32. Speech input goes through femto-asr at 2.39M parameters, so the whole spoken pipeline is 10.1M parameters. It is aimed at cases where the decision has to be made on a small machine.
It scores 0.732 accuracy on the official test split of 400 cases and 2,000 decisions, measured with the typed-decisions scorer. That places it below laya-typed-decisions at 0.767 and Oscar-1-68M at 0.742, and above JEV-9B at 0.706.
The model card lists an apache-2.0 license. The repository is hosted on Hugging Face under igorktech/femto-jev and is tagged for text classification with Safetensors weights.
A recording can be read three ways, all from the audio alone: transcript plus femto-asr hidden frames (joint, the default), transcript only (hyp), or frames only (audio). On SLURP the frames add 13 points to the transcript; on MInDS-14 they add nothing.
It will not beat a 9B-parameter model at reasoning, and a 4.7B audio model hears far better on real speech at about 460 times the size. The frames only help on speech like the training speech, so on new audio conditions expect the transcript number.
No test case is in its training data: 0 of 400 test states. The typed-decisions benchmark is still in-distribution for femto-jev, the Oscar models, and laya-typed-decisions, so the table reads as what each size learned rather than general ability.
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
v-jev 1.4 is a Jev-style decision model on Hugging Face in 0.8B and 4B sizes, built on Qwen3.5 backbones. It returns probabilities for typed questions in one forward pass and adds value extraction and image decisions.
Jev Models & Reimplementations#open-models#classification#calibration
Community
arXiv paper presenting EDU-Jev, a family of single-pass System One decision models at 0.8B, 2B and 4B parameters that return a probability distribution over predefined education answer options without generating any text.
Jev Models & ReimplementationsArticle#arxiv#open-models#classification
Community
MLX builds of Jev-Style-2B-Decision-v4 for Apple silicon: bf16 (3.76 GB) and 8-bit (2.00 GB) folders for text and JSON states, run through one shared jev_style_decision_mlx.py runtime under Apache-2.0.
Jev Models & Reimplementations#open-models#calibration#local
Community
Fine-tuned UltraBERT-380M decision model on Hugging Face that scores a state, question, and option list in one forward pass. Uses sigmoid of yes minus no normalized over options; 72.8% dev and 71.9% test accuracy across 13 datasets, with an evidence head documented on the card.
Jev Models & Reimplementations#open-models#arxiv#calibration
Community
A Hugging Face checkpoint that fine-tunes UltraBERT-380M into a Jev-style decision model: it reads a state, a question and a list of options and picks one in a single forward pass. Reports 72.6% mean accuracy over 13 datasets on the test split.
Jev Models & Reimplementations#open-models#classification#local
Community
Fine-tuned UltraBERT-380M decision model on Hugging Face that reads a state, question, and option list and picks one option in a single forward pass. Scores options with a square-root softplus of yes minus no, normalized over options; 73.2% dev and 71.9% test accuracy over 13 datasets.
Jev Models & Reimplementations#open-models#arxiv#calibration
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.