User repository on GitHub
A 4B decision model: calibrated probabilities for yes/no, choice and score questions over long inputs.
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
Jev directory
CommunityJev Repos & SDKs
Mica
HTTP 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.
- Stars
- 37
- Language
- Python
- License
- APACHE-2.0
- Reading time
- 2 min
- Published by
- Community
- Author
- akivet
- Last verified
- Oct 2026
Used inJev for Software & DevTools
Highlights
- 4B parameters, 32,768 input tokens, 2-255 options, three question types, English/Korean/Chinese and Apache-2.0.
- No text is generated, so a decision costs one read of the input, and several questions about one state share the prefill.
- Docker's first start downloads mica-v0.5-4b-preview-Q8_0.gguf and the tokenizer into the mica-weights volume.
- The image builds for CUDA architectures 86, 89 and 90 (RTX 30xx, RTX 40xx, L40S, H100); --build-arg CUDA_ARCH covers others.
- Non-Docker path: Python 3.10+ with CUDA 12, build llama.cpp via scripts/build_runtime.sh, serve at 127.0.0.1:8010.
Quickstart
bash
git clone https://github.com/akivet/Mica && cd Mica
docker build -t mica .
docker run --gpus all -p 8010:8010 -v mica-weights:/weights micaWatch out
README reviewed 2026-10-17; the excerpt is truncated inside the JSON response example, and this repo is the server for the v0.5 4B Preview weights — benchmarks, file sizes and limits are on the Hugging Face model card, not here.
Questions about this entry
Short answers drawn from the linked project or page.
The HTTP server for Mica v0.5 4B Preview, a small decision model. It runs GGUF weights through llama.cpp and serves POST /v1/systemone: send a state, questions and allowed answers, and get a calibrated probability for each answer.
It speaks the TypeSafe /v1/systemone format, the same SystemOne surface Jev clients use: ask a question about any input, list the answers you allow, and get calibrated probabilities rather than generated text. Several questions about one state share a prefill.
Three: noul (yes/no returned as a probability), choice (one of 2-255 options), and score (ordered levels). The model card lists 4B parameters, 32,768 input tokens and English, Korean and Chinese as main languages.
The Docker image builds for CUDA architectures 86, 89 and 90 — RTX 30xx, RTX 40xx, L40S and H100 — and you can pass --build-arg CUDA_ARCH for others. Without Docker you need Linux, CUDA 12 and Python 3.10 or newer.
Clone the repo, run docker build -t mica . and then docker run --gpus all -p 8010:8010 -v mica-weights:/weights mica. The first start downloads the Q8_0 GGUF and tokenizer into the mica-weights volume.
The README says no text is generated, so the cost of a decision is one read of the input, and several questions about one state share that prefill. The digest does not include v0.5 benchmarks or file sizes — those are on the Hugging Face model card.
The owner is aketiv; the repo has 37 stars, is Apache-2.0 licensed, written in Python, created 2026-09-25 and last pushed 2026-10-12. The weights live at sky7350/Mica-v0.5-4B on Hugging Face.
More Jev resources like this
Serving 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 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 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-cascadePythonAn open-source, non-autoregressive System One decision engine inspired by TypeSafe AI: it evaluates state context against typed questions in a single forward pass, returning probability distributions and calibrated confidence without generating text.
Jev Repos & SDKs#community#github#python
Community
Atulmishra22/jev-litePython
Atulmishra22/jev-litePythonA Python library that turns Jev-style decision models into ordinary function calls: check returns a bool, classify picks one option, label picks every option that applies, and score rates on a scale, with probabilities behind every answer.
Jev Repos & SDKs#community#github#python
Community
jlowin/vibecheckPython
jlowin/vibecheckPythonFrom 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.