CommunityJev Models & Reimplementations
AutoTrust JEV-9B
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.
- 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 40.2M trained parameters — 0.5% of the frozen Qwen3.5-9B backbone — for about 3 B200-hours of training.
- Mean KL of about 0.019 to Jev 1.13's distributions, with choice top-1 agreement 90.2% and ECE 0.0007, all self-reported.
- HumanEval pass@1 stays at 70.7% because System 2 is bit-identical; folding the adapter into the backbone would drop it to 61.6%.
- vLLM computes in bf16, so the card says head bias and temperature must be applied client-side for reproducible decisions.
Watch out
Inputs beyond 1,024 tokens are truncated head 60% / tail 40% at serving unless you raise the limit, --max-model-len 4096 is suggested for decisions, and the same September 2026 vLLM dev build is required. Option-order flips measured 11.5% at 16 options in an independent rerun.
Questions about this entry
Short answers drawn from the linked project or page.
System 2 is the stock Qwen3.5-9B and stays bit-identical, so ordinary chat works through the OpenAI-compatible path. The calibrated decisions live on the jev-decision LoRA module and the /v1/decide route added by the bundled serve script.
The card measured that merge costing HumanEval pass@1 70.7% down to 61.6%, which is why the decision block ships separate and the deployment needs LoRA support on lm_head rather than a single merged checkpoint.
Run vllm serve JEV-9B with --enable-lora --max-lora-rank 32 --lora-modules jev-decision=JEV-9B/adapter_vllm --logprobs-mode processed_logprobs --max-model-len 4096. One engine serves both systems, routing jev-decision requests to System 1.
Apache-2.0. The Qwen3.5-9B backbone is Apache-2.0, and the training corpus SargeDev/jev-distill-corpus-v3 is Apache-2.0 with its openjev_v2 stream additionally CC0.
A single decision takes a median of about 90 ms on one B200, roughly a third of the hosted API's time, and a batched 128 decisions run at about 2.5 ms each, or about 400 decisions per second.
Inputs beyond 1,024 tokens are truncated head 60% / tail 40% at serving unless you raise the limit. The card suggests --max-model-len 4096 for decisions, raised further if System 2 will think at length.
It is an independent student of TypeSafe Jev 1.13 trained on the public corpus, so it mirrors the teacher's mistakes. Mean KL is about 0.019 on 25,376 held-out Jev-labelled rows, and it shares no weights or code with TypeSafe AI.
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
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.
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.
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
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
From the Jev community
Posts from builders shipping with Jev right now.