User repository on GitHub
Small decision models (jev style) for latency critical scenarios
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Models & Reimplementations
Jiwo
Small Jev-style decision models that read a state and typed questions and return a probability for every option in one forward pass without generating text; the 0.75B and 4.2B versions speak TypeSafe's Jev API format at a median of about 50 ms per request on one H100.
- Stars
- 170
- Language
- Python
- License
- MIT
- Reading time
- 2 min
- Published by
- Community
- Author
- jiwidi
- Last verified
- Oct 2026
Works forJev for Classification
Used inJev for Software & DevTools
Highlights
- jiwo-0.8b and jiwo-4b are both Apache-2.0 and listed on the official Decision Index board.
- Request and response format matches TypeSafe's Jev API (POST /v1/systemone); TypeSafe does not endorse it.
- About 50 ms median per request on one H100; 13.1 ms for 0.8b and 31.6 ms for 4b with CUDA graphs on an RTX PRO 6000.
- jiwo-0.8b v2 scores 29.58 public on Decision Index 0.3; untrained Qwen3.5-0.8B scores 6.99 on 0.2.1.
- Needs Python 3.12; training data is mostly English and other languages are not evaluated.
Quickstart
bash
pip install "jiwo @ git+https://github.com/jiwidi/jiwo"Watch out
README reviewed 2026-10-17; both v2 models' Full Decision Index scores were still pending, board ranks come from the 2026-10-07 board, and the README states that non-English languages are not evaluated.
Questions about this entry
Short answers drawn from the linked project or page.
A pair of small decision models in the Jev style for latency-critical scenarios: they read a state and typed questions and return a probability for every option in one forward pass, without generating text.
jiwo-0.8b has 0.75B parameters on a Qwen3.5-0.8B base and jiwo-4b has 4.2B on Qwen3.5-4B. Both are Apache-2.0 and both appear on the official Decision Index board.
The README reports a median of about 50 ms per request on one H100. On one RTX PRO 6000 with CUDA graphs, the median is 13.1 ms for jiwo-0.8b (26.6 ms without graphs) and 31.6 ms for jiwo-4b (34.5 ms without graphs).
It is independent: the README says TypeSafe does not endorse it. Its request and response format is the format of TypeSafe's Jev API (POST /v1/systemone), which is what makes it usable in the same role for latency-sensitive work.
It needs Python 3.12. Install with pip install "jiwo @ git+https://github.com/jiwidi/jiwo", or clone the repo and run uv sync, adding --extra cuda on NVIDIA GPUs for the fast linear-attention kernels.
On the public part of Decision Index 0.3, jiwo-0.8b v2 scores 29.58 and jiwo-4b v2 scores 46.06. Against untrained bases on Decision Index 0.2.1, jiwo-0.8b scores 28.72 versus 6.99 and jiwo-4b scores 45.08 versus 29.20. Full scores were pending at review.
The training data is mostly English and other languages are not evaluated. On CPU, stay in float32 because torch lacks fast bfloat16 matrix products there, and bfloat16 is several times slower on Apple silicon. The v2 Full Decision Index scores were still pending.
More Jev resources like this
An open System One decision model from Interfaze-ai: give lev a state — text, a ticket, an email, or JSON — plus yes/no, choice, or score questions and it answers all of them in one forward pass from logits, returning calibrated probabilities and zero output tokens.
Jev Models & Reimplementations#community#github#python
Community
abhinavexists/levPython
abhinavexists/levPythonA 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-decisionPythonA small, fast decision model — a system one model — for agentic workflows from Strands Labs: it picks between options or rates on a scale faster than an LLM and returns a calibrated confidence with every decision.
Jev Models & Reimplementations#community#github#python
Community
strands-labs/strands-deciderPython
strands-labs/strands-deciderPythonSystem One models from first principles by Qi Li and Xinchao Wang at the National University of Singapore: give Jeva a state and typed questions and it returns candidate probabilities, a Boolean probability, or an expected rubric score, with reusable option memory.
Jev Models & Reimplementations#community#github#python
Community
liqiiiii/jevaPython
liqiiiii/jevaPythonAn open-source, local alternative to Jev: System-One decision models that answer the POST /v1/systemone shape — choice, yes/no, or score — for general decisions and browser-agent steps. Models ship as OpenJev-1.7B, 4B, and 8B, with a pip-installable Python package.
Jev Models & Reimplementations#community#github#python
Community
alanhuangyoo/openjevPython
alanhuangyoo/openjevPythonThree 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/jevaltPythonFrom 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.