User repository on GitHub
Repository for serving models from the Jeff family
- Forks
- 0
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Repos & SDKs
Jeff Serve
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.
- Stars
- 18
- Language
- Python
- License
- APACHE-2.0
- Reading time
- 2 min
- Published by
- Community
- Author
- j-geuter
- Last verified
- Oct 2026
Used inJev for Software & DevTools
Highlights
- Four Jeff 1.0 models served the same way, built from Gemma 4 31B/12B and Qwen3.6-27B / Qwen3.5-4B bases with merged LoRA decision heads.
- Question types: choice (2 to 16 options), noul (yes/no, the answer is P(yes)) and score (ordered levels, 2 to 16); inputs up to 8,192 tokens.
- A decision head turns the last hidden state into one logit per option, and a softmax with a per-question-type temperature gives the probabilities.
- GPU memory spans 80 GB for Jeff-1.0-Large and Qwen-27B, 40 GB for Medium and 16 GB for the 8 GB Qwen-4B model.
- Install with pip, then run jeff-serve with a Hugging Face repo id and a pinned revision; the first start downloads up to 61 GB of weights.
Quickstart
bash
pip install -r requirements.txt && pip install .
jeff-serve jgeuter/Jeff-1.0-Large --revision 275325a5eab52cce4f109a09ee3832f2eb89fe2f --host 127.0.0.1 --port 8013Watch out
README reviewed 2026-10-17; the digest ends mid-sentence just after the first-start download note, so the Docker path, weight licenses and any benchmark numbers were not captured. Model downloads point at Hugging Face repositories under the account jgeuter, while the GitHub owner is j-geuter; pin a tested revision when serving.
Questions about this entry
Short answers drawn from the linked project or page.
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: Jeff-1.0-Large (Gemma 4 31B, 61 GB), Jeff-1.0-Qwen-27B (54 GB), Jeff-1.0-Medium (Gemma 4 12B, 24 GB) and Jeff-1.0-Qwen-4B (8 GB). All four are served the same way.
A decision head turns the last hidden state into one logit per option; a softmax with a temperature fitted per question type (for Jeff-1.0-Large: 1.52 for choice, 1.31 for yes/no and score) gives the probabilities.
A GPU with 80 GB for Large and Qwen-27B, 40 GB for Medium or 16 GB for Qwen-4B. The first start downloads the weights, up to about 61 GB, into the Hugging Face cache or the directory in HF_HOME.
Run pip install -r requirements.txt followed by pip install ., then start jeff-serve with the Hugging Face repo id, the pinned --revision and --host/--port flags; leave --revision out to get the newest version.
noul is a yes/no question whose answer is P(yes); score asks for ordered levels between 2 and 16; choice picks one of 2 to 16 options. A single input can be up to 8,192 tokens.
No. It is an independent open project that builds Jev-style decision models from open LLMs, sharing the typed-question pattern of one probability per option rather than TypeSafe's hosted model.
More Jev resources like this
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.
Jev Repos & SDKs#community#github#python
Community
akivet/MicaPython
akivet/MicaPythonA 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.