User repository on GitHub
Full-weight Qwen decision model with calibrated probabilities, training code, REST API, and playground.
- Forks
- 15
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Models & Reimplementations
AutoJev
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.
- Stars
- 130
- Language
- Python
- License
- MIT
- Published by
- Community
- Last verified
- Oct 2026
About this project
Built with autonomous agents, from research and data generation to training, evaluation, and deployment. The user set the goals and refined the scope; agents executed the work. AutoJev-27B is a multimodal decision model based on Qwen3.8-27B. It returns probabilities over supplied choices in one forward pass per question, with a TypeSafe-compatible API and browser playground.
Quoted from the project's README. Read the full README.
- Languages
- Python
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Reported 84.60% overall accuracy with ECE 0.04282, against Jev's reported 82.79% and ECE 0.05274 on the same chart.
- Built on Qwen3.8-27B with 73,000 unique training examples and 286 updates; the released checkpoint is number 200.
- POST /v1/systemone supports choice, noul, score and optional base64 images; browser playground at localhost:8000.
- Needs Python 3.12, uv and a GPU with room for about 49 GiB of BF16 weights plus runtime overhead.
Watch out
Weights are hosted on Hugging Face and may require authentication while private; image support is not a natural-image accuracy claim. Code is MIT, weights Apache 2.0.
Questions about this entry
Short answers drawn from the linked project or page.
Full-weight SFT over one H200 for 286 updates on 73,000 curated examples; the released checkpoint is number 200, and training code and configs ship in the repository.
A TypeSafe-compatible POST /v1/systemone endpoint answering choice, noul and score questions, plus a browser playground and optional AUTOJEV_API_KEY authentication.
On the project's own chart, AutoJev-27B reports 84.60% overall accuracy with ECE 0.04282 and Brier 0.22027, against 82.79%, 0.05274 and 0.25400 for Jev and 69.83%, 0.06483 and 0.40834 for the Qwen3.8-27B base.
Python 3.12 or newer, uv, and a GPU with room for roughly 49 GiB of BF16 weights plus runtime overhead. Clone the repository, run `uv sync --frozen --python 3.12`, download with `uv run hf download denis-pplx/autojev-27b --local-dir checkpoints/selected`, then `AUTOJEV_CHECKPOINT=checkpoints/selected uv run autojev-serve`. The playground is at localhost:8000 and the API reference at /docs.
The weights live on Hugging Face and are currently private, so authenticate with `uv run hf auth login` before downloading. The code is MIT and the weights are Apache 2.0. It is an independent implementation inspired by Jev, not a distribution of Jev itself.
Training uses cross-entropy, and calibration fits a scalar temperature separately afterward. POST /v1/systemone also accepts optional base64 images, but the published benchmarks measure text decisions, so image support is a capability rather than a natural-image accuracy claim.
The scripts and configs ship and `bash configs/train.sh --help` lists the training arguments, but the exact curated corpus of 73,000 examples is not bundled, so a rerun reproduces the recipe rather than the same data.
Related terms
Glossary definitions related to this entry.
More Jev resources like this
10GitHub 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/rulingPython3GitHub 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/QwevPython77GitHub stars
An Apache-2.0 0.8B decision model that returns calibrated probabilities for yes/no, choice and score questions on CPU. Distributed as GGUF for llama.cpp and served behind a Jev-style /v1/systemone API, so existing Jev clients mostly need only a new base URL.
Jev Models & Reimplementations#community#gguf#local-inference
Community
kouhxp/gutsyHTML
kouhxp/gutsyHTML17GitHub 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/vevPython12GitHub stars
Apache-2.0 ModernBERT fine-tune that answers noul, choice, score and multi-label questions with a probability for every option. It serves Jev's POST /v1/systemone locally so TypeSafe's Python SDK runs unchanged, reporting 0.026 expected calibration error on its test split.
Jev Models & Reimplementations#community#calibration#modernbert
Community
SamratDuttaOfficial/WaterSheepPython
SamratDuttaOfficial/WaterSheepPython5GitHub 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/jebadiahPythonFrom the Jev guides
Original write-ups on what this entry is for.
Base decision checkpoints sit near chance on purpose. The typed-decisions loop, the corpora behind the students, and the temperature fit that turns probabilities into thresholds.
Jev Practices & Patterns#fine-tuning#calibration#open-models
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
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
From the Jev community
Posts from builders shipping with Jev right now.