Organization repository on GitHub
Local typed decisions, contrastive data curation, and model evaluation.
- Forks
- 165
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Models & Reimplementations
Bespoke Nimble
Open recipe for a local Jev-style decision model: a Qwen3.5-9B LoRA that scores one allowed answer token per Choice, boolean or rubric Score question, with data pipeline, training config and eval harness. Reports 90.1% on its 324-example holdout against Jev's 93.2%.
- Stars
- 2.2k
- Language
- Python
- Published by
- Community
- Last verified
- Oct 2026
About this project
Data, Model, Recipe for an open Jev. Model · Updates · Capabilities · Quickstart · Methodology · Documentation and development · Citation. Nimble takes some text and a schema, and makes typed decisions about the text. The schema is the list of questions to answer. Each question is either a choice from a list that you give or a true or false question. For each question, Nimble returns the answer it picked and the probability of each allowed answer.
Quoted from the project's README. Read the full README.
Highlights
- Scores one answer token per field, so there is no generated JSON to parse; output carries the chosen answer and the probability of each allowed answer.
- On 324 held-out examples it matches 90.1% of reference labels, versus 66.4% for its base model Qwen3.5-9B and 93.2% for Jev 1.13.0.
- Latest checkpoint supports an 8,192-token context and up to 255 choices per enum field, with a default temperature of 1.0.
- Runs on Apple Silicon via MLX ParallelScorer or on NVIDIA GPUs via CudaCandidateScorer; needs Python 3.12 and roughly 18 GB of weights.
- Fields are scored independently, so one field cannot see another's answer; your code must check cross-field consistency.
Watch out
The GitHub repository has no detected license file, so code reuse terms are unclear; the released model weights on Hugging Face are Apache-2.0. The authors say it was built in one day with rough edges, and it cannot produce images, text or nested output.
Questions about this entry
Short answers drawn from the linked project or page.
No. The README states the authors did not distill from Jev; the point is to show how to curate data, train and serve a System One-style decision model. Its 90.1% holdout match is measured against reference labels, with Jev 1.13.0 at 93.2%.
It is text-only: no images, no generated text, no nested JSON and no free-text extraction. Each field is scored independently, an enum is limited to 1-255 choices in the latest checkpoint, prompts are capped at 8,192 tokens, and probabilities are not a guarantee of correctness.
Python 3.12, an Apple Silicon Mac where Python itself must run on macOS to reach Metal, or a Linux box with a BF16-capable NVIDIA GPU. The 9B weights are about 18 GB unquantized before runtime overhead, and the LoRA merge step runs on the CPU and needs extra RAM and disk, which is why a 64 GB Mac has more headroom than a 24 GB one. No TypeSafe or generation key is needed for local inference.
Every allowed answer has a one-token code. The scorer reads the model's logits for those codes and turns them into probabilities with a softmax, so there is no generated JSON to parse. On a Mac, ParallelScorer processes the shared context once and scores all fields in parallel; the CUDA scorer scores each field separately with the full prompt.
A temperature was fitted on 22 September 2026 so probabilities better match how often the answers are right. The model picks the same answers as before, but Noul probabilities and Score values do change, so re-test any threshold you compare them against. The latest checkpoint uses T=1.0 and has not had its own fit.
The 2,676 training examples and 324 held-out examples, published on 20 September after being left out of the first release, plus a public benchmark suite over 13 subsets with human labels so you can run Nimble and Jev on the same records. Data curation changes one fact so the correct answer flips, and training is LoRA over the answer tokens only.
The GitHub repository has no detected license file, so terms for reusing the code are unclear rather than permissive. The released model weights on Hugging Face are Apache-2.0. Confirm the code terms before building a product on top of the recipe.
More Jev resources like this
109GitHub stars
Open decision models from a self-improvement loop that publishes every experiment, failures included. They answer Noul, Choice and Score questions on the same POST /v1/systemone wire format as Jev, returning a calibrated probability per option in one forward pass; the 27B scores 65.64 on the Decision Index 0.3 set.
Jev Models & Reimplementations#community#decision-models#system-one
Community
Shanghua-Gao/RSI-JevHTML
Shanghua-Gao/RSI-JevHTML130GitHub stars
An open recipe for a 27B multimodal decision model trained with full-weight SFT on 73,000 examples over one H200, serving choice, noul and score on /v1/systemone with per-checkpoint provenance and released calibration plots.
Jev Models & Reimplementations#community#calibration#qwen
Community
denis-pplx/autojevPython
denis-pplx/autojevPython17GitHub stars
Jev-compatible decision models with vision input: LoRA fine-tunes of Qwen3.5 at 4B and 9B serving Choice, Score and Noul on the /v1/systemone wire format, with screenshots and photos placed beside text in the state. Runs on your own NVIDIA GPU; weights are non-commercial.
Jev Models & Reimplementations#community#vision#open-weights
Community
Xiaooolong/vevPython
Xiaooolong/vevPython10GitHub stars
A local Jev-compatible server that reads Choice, Score and Noul answers from the logits of any MLX checkpoint or OpenAI-compatible endpoint with no training, serving POST /v1/systemone as a drop-in for the official TypeSafe SDK.
Jev Models & Reimplementations#community#mlx#calibration
Community
bradAGI/rulingPython
bradAGI/rulingPython5GitHub stars
Open System One decision models (27B, 9B and 4B on Qwen chat bases, in bf16, GGUF and MLX) that answer Choice, Noul and Score questions with a probability for every option from one forward pass. The repo carries the trainer, data builders, evaluation harnesses and a record of every run.
Jev Models & Reimplementations#community#decision-models#open-weights
Community
getainode/jebadiahPython
getainode/jebadiahPython3GitHub stars
A training-free runtime that turns dense Qwen3 or Qwen3.5 checkpoints into a Jev-style Noul, Choice and Score service, sharing one state prefill across isolated questions and reading each answer from candidate-token logits.
Jev Models & Reimplementations#community#qwen#local-inference
Community
HopLee6/QwevPython
HopLee6/QwevPythonFrom the Jev guides
Original write-ups on what this entry is for.
One model in September, a dozen by October. Clef, Liquid d1, Perplexity, and OpenAI all ship the same typed-answer shape now. How to read the rival benchmarks and pick on your own task.
Jev Models & Reimplementations#decision-model#benchmarks#pricing
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
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.