User repository on GitHub
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Models & Reimplementations
system-one (sahasrarjn)
The Jev architecture rebuilt from Qwen3-0.6B: a calibrated, type-safe decision model on an open backbone. No decoding loop — the state is encoded once, questions are scored in one parallel pass, and answers are a distribution over options declared before the call.
- Stars
- 0
- Language
- Python
- Reading time
- 2 min
- Published by
- Community
- Author
- sahasrarjn
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is System One — iamaamir.
- Languages
- Python
- Capabilities
- Open sourceRuns locally
Highlights
- Jev architecture rebuilt from Qwen3-0.6B: a type-safe decision model on an open backbone with no decoding loop.
- Each question is scored against the encoded state in one parallel pass and returns a distribution over a pre-declared option set.
- Skips the vocabulary projection entirely: it loads AutoModel, not AutoModelForCausalLM, so the 1024 × 151936 matmul is never run.
- Quickstart: pip install -e . then pytest — 9 tests in about 5 seconds, no GPU required.
- Builds training data from the CFPB consumer complaints CSV with no teacher model and no annotation budget.
Quickstart
bash
python -m venv .venv && ./.venv/bin/pip install -e . -r requirements.txt
./.venv/bin/python -m pytest # 9 tests, ~5s, no GPUWatch out
README and GitHub metadata reviewed 2026-10-17; the digest ends inside the how-it-works explanation, so training results and calibration numbers are unavailable, and no licence is listed. Created 2026-09-19 and pushed 2026-09-25 with 0 stars.
Questions about this entry
Short answers drawn from the linked project or page.
An open reimplementation of the Jev architecture built from Qwen3-0.6B: a calibrated, type-safe decision model that returns a distribution over declared options instead of generating text.
The state is encoded once and every question is scored against it in one parallel pass; an answer outside the declared option set is not unlikely, it is unrepresentable.
The quickstart creates a venv and installs with pip install -e . plus requirements, then runs pytest — 9 tests in about five seconds without a GPU.
The README builds data from the CFPB consumer complaints CSV, described as requiring no GPU, no teacher model and no annotation budget, via scripts/build_data.py with --civil-n 40000.
It never builds the vocabulary projection, so for Qwen3-0.6B the 1024 × 151936 lm_head matmul that a decoder pays at every position is skipped.
No. The README calls it the Jev architecture rebuilt from Qwen3-0.6B on an open backbone, so it is an independent implementation of the same idea rather than the hosted model.
A Python repository with 0 stars, created 2026-09-19 and pushed 2026-09-25 per GitHub metadata, with no licence listed in the digest.
More Jev resources like this
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/gutsyHTMLA Chinese-language decision model that returns a probability for each fixed candidate option in one forward pass, with no generated text or chain of thought. It is a LoRA fine-tune of Qwen3.5-0.8B on 13,048,000 CJ-Dataset rows, benchmarked on CJ-Bench against Chinese-Jev.
Jev Models & Reimplementations#community#github#python
Community
yanqiangmiffy/chinese-decisionPython
yanqiangmiffy/chinese-decisionPythonOpen recipe for calibrated decision models: an LLM teacher writes training data from a one-sentence task description, a LoRA plus decision head on Qwen3-4B-Base answers Choice, Score and Noul questions with calibrated probabilities, evaluated next to Jev on identical items.
Jev Models & Reimplementations#community#calibration#training
Community
scienthoon/lucePython
scienthoon/lucePythonThree open-weights 4B decision models for English (Deem-4B), Turkish (Karar-4B) and German (Wähler-4B) that choose between options and return a probability for each. The server implements the Jev API, POST /v1/systemone, so existing Jev clients work by changing the endpoint.
Jev Models & Reimplementations#community#github#python
Community
mertkayacs/jevaltPython
mertkayacs/jevaltPythonA small from-scratch alternative to Jev and Laya-style System 1 decision models: a roughly 22M-parameter shared encoder trained with per-example calibrated confidence labels in PyTorch and exported to ONNX for local, sub-second inference.
Jev Models & Reimplementations#community#github#onnx
The typed-decisions checkpoint of the Laya family on ModelScope: a 421M ModernBERT-large fine-tune reporting 0.766 accuracy over 2,000 decisions, above TypeSafe Jev 1.13.0's published 0.727 and the 0.735 teacher ceiling, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
From 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.