What actually changed to make the GPU code run on CPU?

Upstream pins one accelerator in the loader while downstream scoring follows the model device and skips accelerator sync elsewhere. This port swaps only the loader to CPU float32 and reuses the original scoring code.

How is a decision read without generating any text?

Each option becomes one letter in a single chat turn ending where the answer goes, each letter is verified as one clean token, and softmax runs over those slots alone from one forward pass.

How does shared mode avoid paying prefill per question?

When criteria judge the same state, the state prefills once into native cache, replicates across branches, and evaluates all option positions in one batched forward, so prefill cost is paid once.

Should I trust the loan approve output when risk is high?

The demo flags high risk yet leans approve on the paired decision, which the docs call a small-model slip that larger models resolve. Treat paired fields as needing joint review rather than trusting each alone.

What model does JEV-CPU run and at what speed?

Qwen/Qwen3-0.6B in float32 (~2.4 GB) on CPU, with each decision about 1 s and model load 5-17 s; no GPU, no waitlist and no text generated.

What hardware and Python does JEV-CPU need?

Python 3.10+ and about 3 GB of RAM, no GPU; install the CPU PyTorch wheel plus transformers and accelerate, then run the CLI or the stdlib web server on port 8080.

What API does JEV-CPU's web server expose?

GET /api/health and POST /api/decide, where decide takes a state and a list of criteria, each criterion having an id, a question and options with ids and descriptions.

CommunityJev Demos & Experiments

JEV-CPU

CPU port of the SemIf decision pattern with a browser UI: type state and runtime criteria, then read typed option probabilities from one open-model forward pass with no GPU.

See highlights

Open github.com

Stars: 18
Language: Python
License: MIT
Published by: Community
Author: leesk212
Last verified: Oct 2026

leesk212/JEV-CPU

User repository on GitHub

View on GitHub

Run SemIf (Jev-style semantic-if decisions) on a CPU — no GPU. Reads typed option probabilities straight from an open model in one forward pass, plus a web UI.

Forks: 4
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

Semantic ifs from open models — on a laptop CPU, no GPU. A CPU port of SemIf (formerly OpenJev), with a web UI. 📄 Read the full technical report → · Run it locally · How it works · Web UI. Live PoC — the state is typed in, criteria are added, and the decision is read from Qwen3-0.6B's option logits in 1 s (no text generated). One CPU engine across eight domains: support, content moderation, code-review triage, incident severity, email intent, compliance, loan/credit risk, and support prioritization.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import

Works forJev for Classification

Used inJev for Software & DevTools

Languages: Python

Highlights

Loads a small open model on CPU in float32 and reuses it across requests with no accelerator required.
Exposes a three-pane web UI plus a small JSON API for health checks and batched criteria decisions.
Reports per-option logits and probabilities in about a second per decision without any decoding loop.
Documents a cross-field slip on the loan demo and points to larger models for consistent paired judgments.

Watch out

Small-model judgments can be inconsistent across related fields, and the same scoring code simply runs without accelerator sync on CPU. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

open-jev-typed-decision-engine

Open 150M encoder that reproduces the typed noul, choice, and score interface in one forward pass, trainable on a free GPU and runnable locally for nothing.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

intikhab49/open-jev-typed-decision-enginePython

TypeLLM

941

Library that adds schema-guaranteed typed outputs to existing autoregressive LLMs without changing weights, served through SGLang or a hosted API.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

TypeLLM/TypeLLMPython

solar-mini4-jev

Drop-in service that exposes Upstage Solar models through the System One shape, accepting state plus noul, choice, or score questions.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

hunkim/solar-mini4-jevPython

open-medical-jev

Research recipe that fuses two frozen open readers into Jev-style yes/no judgments with a fit-free router, confidence gate, and candidate sets.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

FeiLiuEM/open-medical-jevPython

jev-robot-control

Simulation study that drives one xArm7 apple-to-plate task with Jev, a large reasoning model, and a small model through identical incremental controls.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

openroboto-ai/jev-robot-controlPython

OpenJev-Vision

Open research toolkit that encodes an image once and answers multiple typed choice, noul, and score questions from a shared distribution, plus a local 0.6B text-decision model.

Jev Demos & Experiments#community#jevbest#catalog-import

Community

IamBusy/OpenJev-VisionPython

From the Jev guides

Original write-ups on what this entry is for.

Per-Tick Decisions in Games

Sep 28, 2026

A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.

Jev Demos & Experiments#gaming#agents#control

Read article

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai