JevEmbed
Toolkit for reading Choice, Score and Noul decisions out of embedding models, with data, fine-tuning, benchmarks and a server.
Overview
JevEmbed from HIT-TMG turns embedding models into decision models. It is a one-stop toolkit covering supervised data synthesis for Choice, Score and Noul with configurable label quotas, fine-tuning, inference, a versioned benchmark module with unified evaluation and structured results, and an interactive playground with model comparison and a Game Lab of model-controlled pixel games.
Released models include JevEmbed-Qwen3-Embedding-4B and 0.6B and JevEmbed-KaLM-Embedding-V2.5, with merged weights and LoRA adapters on Hugging Face, plus support for CLM-v0.1-8B. JevEmbed-Data provides 1.67 million labelled Choice, Score and Noul questions for fine-tuning. Decisions can be run from a Python API, a CLI with model configs, or an HTTP server that accepts Jev-style requests on /v1/systemone, so existing clients can call an embedding-based decision model. Core imports and an explain mode need no model weights, and Python 3.10 to 3.12 is recommended for the tested model stack.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which option best matches the state? (e.g. how to handle a wrong-size shoe delivery)(inferred) | — |
| Other | Score and Noul decisions read from embedding similarity | — |
How it works
- Synthesise labelled decision data
- Fine-tune an embedding model
- Serve via Python, CLI or /v1/systemone
- Benchmark and explore in the playground
Strengths
- Uses compact embedding models as decision models
- Large open training dataset
- Jev-style HTTP endpoint
Caveats
- Accuracy relative to hosted Jev not stated in the excerpt
- Tested on specific Python versions
Getting started
python -m pip install -r requirements.txt && python -m pip install -e . --no-deps, then `python -m jevembed --config configs/jevembed-qwen3-embedding-0.6b.yaml --serve`.At a glance
- Type
- CLI
- Status
- research
- Language
- Python
- Runtime
- Library / CLI / HTTP server
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Meet JevEmbed — turn embeddings into decisions.
wh000wh000/awesome-jev-liveMeet JevEmbed — turn embeddings into decisions. Choose, score, and judge with your choice of embedding model.
hellogumbo/awesome-jevChoice, score and noul decisions read out of an embedding model of your choosing, with a Python API, a CLI and an optional HTTP server.
MrJev/awesome-jev
Frequently asked questions
How can an embedding model make decisions?
JevEmbed reads Choice, Score and Noul answers from embeddings, optionally after fine-tuning.
Is there training data?
Yes, JevEmbed-Data has 1.67 million labelled questions.
Can Jev clients use it?
Yes, its server accepts requests on /v1/systemone.
Related projects
llama.cpp-based C++ server that turns any LLM into a System One decision model, reading answers from next-token logits.
ProjectbetaC++MLX library that turns an existing local model into a decision engine by reading candidate logits, with no fine-tuning, on Apple Silicon.
ProjectprototypePythonLocal server (xks) implementing Jev's /v1/systemone contract on an open-weight LLM, accelerated with Xeon Phi cards.
ProjectresearchRustEducational C11 library and CLI that turns DiffusionGemma 26B into a CPU-first OpenJev-style decision engine reading candidate logits.
ProjectresearchCExperimental engine that makes an existing LLM do Jev-like classification and scoring by reusing context and building JSON.
ProjectresearchPythonReads typed decisions from one forward pass of PrismML's 1.75-bit Bonsai 27B under llama.cpp, using the OpenJev/SemIf logit method.
ProjectresearchPython