CommunityJev Models & Reimplementations
AutoTrust JEV-27B
An 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.
- Published by
- Community
- Author
- AutoTrust AI Lab
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Only 108.9M of 27B parameters are trained: a rank-16 LoRA plus a 24-slot decision head, on a frozen Qwen3.8-27B backbone.
- The decision head starts from the backbone's own lm_head rows, so at step zero it already matches the pretrained restricted distribution.
- Six-group decision mean 84.07 against TypeSafe Jev 1.13's 83.85 in AutoTrust's own full runs on both models.
- Mean KL of about 0.017 to Jev's distributions on 25,376 labelled rows, with noul AUROC 0.995.
- Ships POST /v1/decide while ordinary chat completions remain the untouched System 2 backbone.
Quickstart
bash
curl -s localhost:8000/v1/decide -H 'Content-Type: application/json' -d '{
"kind": "noul",
"state": "The deploy failed twice and customers see 500s.",
"question": "Does this need attention right now?",
"options": ["no", "yes"]
}'Watch out
One 80 GB GPU gets you started — about 52 GB of weights plus KV cache, with --max-model-len 131072 suggested — and it needs a September 2026 vLLM dev build. It inherits the teacher's known weaknesses in counting, dates, arithmetic, and adversarial inputs, and the card says it is not for high-stakes decisions.
Questions about this entry
Short answers drawn from the linked project or page.
Not natively. The bundled serve_decide.py adds POST /v1/decide with {kind, state, question, options} returning a calibrated probability per option, beside standard OpenAI-compatible endpoints for System 2 chat. The card lists the six option kinds and limits on GET /v1/decide/info.
AutoTrust reports mean KL of about 0.017 to Jev 1.13's distributions, choice top-1 agreement of 90.5%, and noul AUROC 0.995 on held-out rows — all self-reported. An independent human-labelled run cited on the card puts JEV-27B at 96% of the teacher's accuracy with 16 options.
The card measured that merging an adapter into the backbone cost the comparable JEV-9B 61.6% versus 70.7% HumanEval pass@1, which is why the System 1 block stays separate and the serving path needs LoRA support on lm_head rather than a merged checkpoint.
About 54 GB of weights and one 80 GB GPU is the starting point, with --max-model-len 131072 suggested. The card requires a vLLM build with qwen3_5 support, LoRA on lm_head, --logprobs-mode and allowed_token_ids, tested against a September 2026 development build; start-up takes 3 to 8 minutes because of CUDA-graph capture with LoRA enabled.
Apache-2.0, with Qwen/Qwen3.8-27B declared as the base model and SargeDev/jev-distill-corpus-v3 as the training corpus, also Apache-2.0. It is an independent open-weights student built by AutoTrust AI and shares no weights or code with TypeSafe, which makes it easy to confuse with the unrelated AutoJev-27B full-weight SFT model.
A choice question takes 2 to 256 options with no retraining: the first 16 are labelled A through P and the rest continue with Q through Z and two-letter labels, with the response's adaptation field saying native or wide-labels. On CLINC150 with all 150 intents it scores 89.5% from names alone and 93.8% with a one-line description each. The backbone's native context is 262,144 tokens.
JEV-9B gives nearly the same answers 2.6x faster, around 90 ms against 137 ms on one B200, and is the right pick for routing, moderation and short option lists. Use JEV-27B for option lists beyond about eight, unfamiliar task families, code-rule checks, fraud screening, or whenever the System 2 path matters.
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 NewsRelated terms
Glossary definitions related to this entry.
More Jev resources like this
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
GGUF builds of the Jev-Style v3 decision model on Qwen3.5-0.8B: a 0.53 GB Q4_K_M file the card says is decision-identical to FP32 on 240/240 parity rows, served through a pinned llama.cpp scorer.
Jev Models & Reimplementations#community#open-models#local
Community
A 400M-parameter Jev-compatible decision model with a browser build: 0.796 on typed decisions against Laya's 0.766, 17 ms per question on an L40S, and an ONNX file that runs in a tab.
Jev Models & Reimplementations#community#open-models#local
Community
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++8GitHub 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/verdictPythonFrom 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
Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.
Jev Tools & CLIs#open-models#local#onnx
Read article
From the Jev community
Posts from builders shipping with Jev right now.