CommunityJev Models & Reimplementations
OpenDecider Nano
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.
- Published by
- Community
- Author
- Manjunath Janardhan
- Last verified
- Oct 2026
Used inJev for Software & DevTools
Highlights
- 0.796 fine-tuned on typed decisions against Laya's 0.766, with the card's own confidence interval of +0.014 to +0.044.
- Runs in a browser at 47 ms per question on WebGPU from a 450 MiB ONNX file that matches the benchmark answers on at least 99.5% of questions.
- opendecider serve exposes /v1/systemone with dynamic batching, auth, Prometheus metrics, and a Docker image.
- ECE 0.092 on general decisions, ahead of Jev's 0.164 and Laya's 0.327 in the card's table.
Watch out
English-only, and the card concedes Laya wins on its own trained datasets (spam 0.99, phishing 0.98) while Jev leads general decisions (0.730 against 0.680). Phishing is its weakest task at 0.63, terse option labels hurt, and the distillation sources are Qwen3-235B and DeepSeek V4.1 Flash.
Questions about this entry
Short answers drawn from the linked project or page.
Yes for the shape: opendecider serve exposes Jev's POST /v1/systemone, with an MCP server and a browser build beside it. The card's comparisons are self-run — the Laya numbers come from a fine-tuned-versus-fine-tuned table it measured itself.
The card lists a Mac with 8 GB, an NVIDIA GPU with 4 GB, or CPU alone at 0.1–0.7 s per question, tested on a 16 GB M4 Mac mini at 28 ms. Weights are stored bf16 at 0.8 GB and run in fp32 at about 2 GiB.
opendecider serve --model manjunathshiva/opendecider-nano exposes a Jev-compatible POST /v1/systemone with dynamic batching, back-pressure, auth, Prometheus metrics and Docker images. It was load-tested at 100 concurrent users with zero errors, serving 50 requests/s on one NVIDIA L4 and 24 on eight CPU cores.
pip install opendecider, then load('manjunathshiva/opendecider-nano'), which downloads about 0.8 GB on first use, and call model.system_one(state, questions) with choice, score and noul questions to get a calibrated probability per option.
Phishing is its weakest task at 0.63, terse option labels hurt, and it is English-only. Laya wins on its own trained datasets (spam 0.99, phishing 0.98) while Jev leads general decisions (0.730 against 0.680). Its distillation sources are Qwen3-235B and DeepSeek V4.1 Flash.
It reports ECE 0.092 on general decisions, ahead of Jev's 0.164 and Laya's 0.327 in the card's table.
Yes. @opendecider/web runs the ONNX build in a browser or in Node, Bun and Deno, with an fp16 build scoring 40 questions at once in 1.1 seconds on WebGPU, and opendecider.guard blocks jailbreaks and prompt injection in each agent framework's hook.
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
A browser WebGPU export of Julia 1: the same 551 MB decision checkpoint, a Rust WASM tokenizer, and a JavaScript API that answers typed questions client-side with no server or key.
Jev Models & Reimplementations#community#open-models#local
Community
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
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
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++From 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.