User repository on GitHub
Experimental typed decisions from causal LMs: shared-prefix inference, calibration, compiled heads, and explicit fallback routing
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Repos & SDKs
System One (experimental runtime)
An experimental semantic decision runtime for ordinary Hugging Face causal language models: it returns typed Boolean or Choice distributions straight from logits, reuses context prefill across fields and can compile repeated fields into small classifiers.
- Stars
- 0
- Language
- Python
- Reading time
- 2 min
- Published by
- Community
- Author
- Luke458
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is System One — iamaamir.
Used inJev for Software & DevTools
- Languages
- Python
- Capabilities
- Open sourceRuns locally
Highlights
- Returns typed Boolean and Choice distributions directly from logits instead of generating text through a chat template.
- Reuses context prefill across fields and can compile repeated fields into lightweight classifiers over a frozen model's hidden states.
- MiniCPM5-2B is the first full checkpoint tested; the README says the original zero-shot method is prompt-sensitive.
- All reported quality measurements are small synthetic experiments, per the README's own caveat, and no CUDA-only extensions are required.
- Tested GPU runtime is an RX 9070 XT with Torch 2.13.0+rocm10.0.0, Transformers 5.8.0 and bfloat16; Python 3.11+ is required.
Quickstart
bash
python -m venv .venv
. .venv/bin/activate
pip install torch --index-url https://download.pytorch.org/whl/cpu
pip install -e '.[test]'Watch out
README reviewed 2026-10-17; the excerpt truncates in the setup-script notes and no license is listed in the metadata. The README itself says the method is prompt-sensitive and all quality measurements are small synthetic experiments, so no production claims are supported.
Questions about this entry
Short answers drawn from the linked project or page.
An experimental semantic decision runtime for ordinary Hugging Face causal language models: it returns typed Boolean or Choice distributions directly from a model's logits rather than generating text.
It does not claim any TypeSafe link: it implements typed decisions from ordinary causal LMs, and the README names MiniCPM5-2B as the first full checkpoint tested. It is an independent take on the same typed-decision idea.
Python 3.11+ is required, with development on 3.12. Create and activate a virtual environment, install the PyTorch build for your CPU or GPU, then install with pip install -e '.[test]' and run the tests with OMP_NUM_THREADS=4 python -m pytest -q.
CPU and ROCm paths are documented; the tested GPU runtime is an RX 9070 XT with Torch 2.13.0+rocm10.0.0 and Transformers 5.8.0 in bfloat16. The README says no CUDA-only extensions are required.
No key is involved: the runtime works with ordinary Hugging Face checkpoints. The virtual environment, model weights and fitted head artifacts are excluded from Git, so you download or fit those yourself.
The README says the original zero-shot method is prompt-sensitive and that all reported quality measurements are small synthetic experiments, so the results should not be read as evidence of production readiness.
It is a personal Python project by GitHub user Luke458, created and pushed on 2026-09-16 with zero stars at sweep time, and no license is listed in the repository metadata.
More Jev resources like this
A System One layout optimizer for decision-model workloads: it reorders rows and fields so a prefix-caching backend reuses input computation, then returns per-record decisions in original order. Its bundled backend serves AutoTrust JEV-9B through vLLM.
Jev Repos & SDKs#community#github#python
Community
0814wdwd/solo_jevPython
0814wdwd/solo_jevPythonA Bangla-focused adaptation of Laya, the non-autoregressive typed-decision model: targeted data preparation, fine-tuning and evaluation for classification, routing and scoring on Bangla, Banglish and code-switched text, with the checkpoint still to come.
Jev Repos & SDKs#community#github#python
Community
shamspias/laya-banglaPython
shamspias/laya-banglaPythonHTTP server for the Mica v0.5 4B Preview decision model: GGUF weights run through llama.cpp and answer noul, choice and score questions with calibrated probabilities over the TypeSafe /v1/systemone format.
Jev Repos & SDKs#community#github#python
Community
akivet/MicaPython
akivet/MicaPythonServing code for the Jeff 1.0 family of Jev-style decision models: a state, a typed question and its options go in, and one forward pass returns a probability for each option. Four open models from 4B to 31B parameters are served the same way.
Jev Repos & SDKs#community#github#python
Community
j-geuter/jeff-servePython
j-geuter/jeff-servePythonA frozen decision server that serves Gemma 4 31B — quantized to FP8 at startup — in TypeSafe's /v1/systemone format, reading the next-token distribution instead of generating tokens. It needs one FP8 NVIDIA GPU with 80 GB and has no authentication.
Jev Repos & SDKs#community#github#python
Community
krishna-gogineni-765/deck31bPython
krishna-gogineni-765/deck31bPythonAn open-source, locally executed decision engine in the spirit of Jev (TypeSafe AI): state becomes features, and Noul, Choice and Score heads return typed, calibrated probabilities without generating text. It ships a cheap-to-strong cascade and a full train-calibrate-evaluate-deploy cycle.
Jev Repos & SDKs#community#github#python
Community
slavabars/decision-cascadePython
slavabars/decision-cascadePythonFrom 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.