User repository on GitHub
Dex: open-source, Jev-compatible, non-autoregressive System 1 decision model on an Ettin-1B encoder. Apache 2.0.
- Forks
- 1
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Models & Reimplementations
Dex
An open-source, Jev-compatible System One model: Dex answers choice, score and yes/no questions over text in a single forward pass from an Ettin-1B encoder, and speaks the same /v1/systemone shape as Jev.
- Stars
- 1
- Language
- Python
- License
- APACHE-2.0
- Reading time
- 3 min
- Published by
- Community
- Author
- CanadaOrNaw
- Last verified
- Oct 2026
Used inJev for Software & DevTools
Highlights
- Non-autoregressive: one forward pass produces typed answers, never generated text, so there is nothing to parse and nothing to hallucinate.
- Built on Ettin-1B, a bidirectional encoder (1.1B parameters, 784M fine-tuned) with a decision head trained from scratch against strictly proper scoring rules.
- Speaks the same POST /v1/systemone request and response shape as Jev; point an existing client at http://127.0.0.1:18766/v1/systemone and drop the key.
- Scored 71.4% against human labels across 5,327 questions, closing about 40% of the gap between stock Laya at 64.0% and the JEV API at 81.8%.
- Runs on an 8 GB card: 2.4 GB of weights in bfloat16, 18 ms for one short-text question versus 886 ms for the JEV API on the same machine.
Quickstart
bash
git clone https://github.com/CanadaOrNaw/dex && cd dex
python -m venv .venv && . .venv/bin/activate && pip install -r requirements.txt
huggingface-cli download CanadaOrNaw/dex-1b --local-dir model
python server/dex_server.py 18766Watch out
Grounded in the truncated repository README and GitHub metadata reviewed on 2026-10-15; the benchmark table is the author's own measurement on one RTX 3070 Ti scored against human labels, the README stops inside the Decision Index section, and differences from Jev include no credits_remaining_usd field and one request at a time.
Questions about this entry
Short answers drawn from the linked project or page.
Dex is an open-source, Jev-compatible, non-autoregressive System One decision model under Apache 2.0. It answers typed choice, score and yes/no questions over any text in a single forward pass and never generates text, so it runs locally and for free instead of calling a hosted API.
It uses the same POST /v1/systemone request and response shape as TypeSafe Jev, with answer fields exactly matching Jev's for each question type (choice, score and noul), checked against real Jev responses. You can point an existing Jev client at http://127.0.0.1:18766/v1/systemone and drop the key; /api/v1/decide and /v1/decide are also accepted.
Clone the repository, create a Python virtual environment, install requirements.txt, download the dex-1b weights with huggingface-cli (2.4 GB) and run python server/dex_server.py 18766. The repo also ships examples/curl.sh for a sample call and tests/test_server.py with 12 checks against the running server.
Against original human labels across 5,327 questions: JEV 81.8%, Dex 71.4% and stock Laya 64.0%. On the ten most-cited JEV use cases Dex reaches 81.5% versus Jev's 84.6% and Laya's 68.2%; on tasks it was never trained on it scores 74.4% versus 84.5% and 63.0%.
No API key or fees: it runs locally. Weights are 2.4 GB in bfloat16, measured at 18 ms for one short-text question on an RTX 3070 Ti with 8 GB. The README also shows a CPU loading path that keeps the encoder in bfloat16 before moving it to GPU, and ships a systemd unit plus a CPU Dockerfile.
The response omits credits_remaining_usd, no key is required unless you set DEX_TOKEN to demand a bearer token, and one request is handled at a time. Answers are produced locally, so usage reports input_tokens with cost_usd at zero.
CanadaOrNaw released Dex under Apache 2.0. It wraps an Ettin-1B bidirectional encoder (1.1B parameters, 784M fine-tuned) with a decision head trained from scratch against strictly proper scoring rules, and it places 11 points ahead of stock Laya on tasks it was never trained on.
More Jev resources like this
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/vevPythonA 144.3M-parameter bilingual typed decision model fine-tuned from jhu-clsp/mmBERT-small: it answers noul, choice, and score questions in one forward pass with calibrated confidence and zero output tokens, and it runs on CPU.
Jev Models & Reimplementations#community#catalog-import#cobanov
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-JevHTMLA 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/rulingPythonOpen 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/jebadiahPythonA 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.
From the Jev AI model community
Posts from builders shipping with Jev right now.