CommunityJev Models & Reimplementations
AutoTrust JEV-27B-VL
The vision sibling: the same text-trained System 1 head on Qwen3.8-27B's image encoder, taking images zero-shot, ranked first of 20 vision decision models on the community Jev Decision Index 0.3.
- Published by
- Community
- Author
- AutoTrust AI Lab
- Last verified
- Oct 2026
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- System 1 was trained on text only; every image decision is zero-shot on the untouched vision encoder.
- Full score 69.82 with ECE 0.019 on the Vision board of Jev Decision Index 0.3, above the stock 397B reference by 6.4 points.
- Text parity with JEV-27B: mean probability difference of 0.010 on 1,000 answer-checking decisions.
- --max-num-seqs 8 is mandatory on the vLLM path; more than eight sequences returns wrong System 1 probabilities.
Watch out
Needs one 80 GB or larger GPU and the same vLLM dev build as JEV-27B. Image calibration has not been measured systematically, and the computer-use demo scored 95% of 60 tasks only with numbered boxes plus element text — with boxes alone it dropped to 10%.
Questions about this entry
Short answers drawn from the linked project or page.
No. The System 1 adapter and decision head were trained on text only, and the vision encoder is the stock Qwen3.8-27B module, so every image decision is zero-shot and the card says its calibration on image tasks has not been measured systematically.
First of 20 vision decision models on the community Jev Decision Index 0.3 with a full score of 69.82 and ECE 0.019, above the runner-up's 67.66 and 6.4 points ahead of the stock Qwen3.5-397B-A17B reference. The board is community-run rather than AutoTrust's own.
bash serve.sh runs a vLLM server on :8000 with a POST /v1/decide route. It needs one GPU with 80 GB or more and --max-num-seqs 8, which is mandatory because more than eight sequences in a batch returns wrong System 1 probabilities.
Apache-2.0. The repository holds Qwen/Qwen3.8-27B's weights unchanged plus the JEV System 1 adapter and decision head from autotrust/JEV-27B.
System 1 completes 95% of 60 computer-use tasks at about 0.26 s per click, but only 10% with numbered boxes alone, and 75% of 20 robot-arm pick-and-place scenes, the best of the JEV family.
The backbone's native context is 262,144 tokens. In a long-text test all 10 yes/no and 10 sixteen-option decisions were correct at every length tested up to 250K, with median latency rising to 28.7 s.
Yes: on 1,000 answer-checking decisions the mean probability difference between the two is 0.010 and 99.8% fall on the same side of 0.5. The System 1 adapter and decision head were trained on text only.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
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 NewsHacker News: A one-token logprob wrapper that also classifies webcam frames.
allanrbo151 pointsA single function Jev-like wrapper for LLMs, including vision modelsRead the thread on Hacker NewsMore Jev resources like this
1.3kGitHub stars
A local yes/no decision model that speaks Jev's wire format on a laptop CPU, refusing Choice and Score until those question types ship.
Jev Models & Reimplementations#community#python#open-models
Community
feder-cr/jevC++
feder-cr/jevC++58GitHub stars
An Apache-2.0 decision engine with a TypeSafe-compatible endpoint, document answers with source passages, and a ViZDoom demo that picks actions from structured game state.
Jev Models & Reimplementations#community#python#local
Community
deepanwadhwa/OpenDecisionPython
deepanwadhwa/OpenDecisionPython8GitHub stars
A Python server in front of llama-server that reads one-token label probabilities and exposes them as POST /v1/systemone.
Jev Models & Reimplementations#community#python#open-models
Community
khimaros/verdictPython
khimaros/verdictPythonAn open distillation of Jev 1.13 onto Qwen3.8-27B: a 108.9M-parameter decision block answers typed questions in one prefill pass and adds a POST /v1/decide route beside a full System 2.
Jev Models & Reimplementations#community#open-models#local
Community
The 9B member of AutoTrust's Jev students: a 40.2M-parameter decision block on Qwen3.5-9B with 0.019 mean KL to Jev 1.13 and 70.7% HumanEval left untouched, small enough for one modest GPU.
Jev Models & Reimplementations#community#open-models#local
Community
AutoTrust's Gemma-based decision model with adaptive thinking: System 1 answers below 0.8 confidence hand off to the full Gemma 4 26B and mix the two by half, behind POST /v1/decide.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
Interface, size, license, claims — in that order. A selection guide across the system-one catalog, from a 706K form scorer to 27B Jev students.
Jev Practices & Patterns#open-models#evaluation#benchmarks
Read article
Hundreds of repositories carry the system-one tag, but the ecosystem is a dozen ideas in different sizes. A tour of the families, the licenses, and what a download does not include.
Jev Models & Reimplementations#open-models#hugging-face#local
Read article
Jev takes text only; Clef-omni adds image, audio, and video input and Liquid d1 adds images, each keeping the same typed-answer shape. Where that matters, where it does not, and the costs the vendors underplay.
Jev Models & Reimplementations#multimodal#open-models#decision-model
Read article
From the Jev community
Posts from builders shipping with Jev right now.