User repository on GitHub
- Forks
- 7
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Demos & Experiments
Solar Mini4 Jev
Drop-in service that exposes Upstage Solar models through the System One shape, accepting state plus noul, choice, or score questions.
- Stars
- 45
- Language
- Python
- Published by
- Community
- Author
- hunkim
- Last verified
- Oct 2026
About this project
A drop-in wrapper that exposes Upstage Solar Pro4 (default; solar-mini4 optional) through the TypeSafe Jev System One API shape. Supported question types: noul | choice | score (same schema as Jev). Hosted BYOK endpoint — no deploy needed to try it: Bring your own Upstage API key. The server does not store it. \ Disclaimer: results use a subset of public Jev benchmarks. The live API is Bring Your Own Key. The server never stores your Upstage key. POST /v1/systemone — System One API (BYOK)
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Serves POST /v1/systemone with the same question schema as Jev, translating decisions into lettered-option prompts over Solar chat.
- Defaults to the stronger Pro model with the smaller Mini model selectable, and always answers with reasoning disabled as a single option.
- Runs as a bring-your-own-key hosted endpoint or a local FastAPI app, with health, scorecard, docs, and machine-readable reference pages.
- Includes comparison scripts and a benchmark table over public Jev benchmark subsets, marked as a subset rather than a full evaluation.
Watch out
Bring your own Upstage key; the hosted server does not store it. Benchmarks cover only a subset of public Jev benchmarks from one region. 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.
It is a compatible wrapper, not the hosted model: it translates System One questions into lettered-option Solar chat prompts and returns the same answer shape, so compare behavior on your own tasks.
The default stronger model leads the subset table while the smaller Mini model answers faster; the base model is selectable by environment, with reasoning always disabled to a single option letter.
Send a JSON POST to the hosted endpoint with your Upstage key in the header and a state plus questions body; no deploy is needed, and health and docs pages show the shape.
No. Treat it as a subset datapoint with regional latency effects and provider changes over time; rerun the included comparison scripts with your key before choosing.
Locally it is a FastAPI app: pip install -r requirements.txt then uvicorn server:app on a port of your choice. For Vercel, npx vercel@latest --prod, and no UPSTAGE_API_KEY is required in project env for production because the endpoint is bring-your-own-key. Routes are GET / for the scorecard, GET /health for liveness, and POST /v1/systemone. Auth is either X-Upstage-Api-Key or an Authorization: Bearer header.
UPSTAGE_API_KEY is required for real calls. SOLAR_MINI_MODEL picks the base model and defaults to solar-pro4, which currently resolves to solar-pro4-260806; solar-mini4 is faster but less accurate. TYPESAFE_API_KEY is only needed to call the real Jev from jev_ref.py during benchmarks, on jev-latest. Reasoning is always none, so the model answers with a single option letter.
engine.py rewrites a Jev System One request as a lettered-option prompt against Solar chat, with reasoning forced off so the reply is a single option letter that maps back onto the answer shape. server.py exposes that as POST /v1/systemone, and jev_ref.py is a separate client that calls real Jev purely so the comparison scripts can produce the benchmark table. The repo is experimental research code and says not to commit API keys.
More Jev resources like this
46GitHub stars
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
941GitHub stars
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
TypeLLM/TypeLLMPython38GitHub stars
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
FeiLiuEM/open-medical-jevPython62GitHub stars
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
40GitHub stars
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
IamBusy/OpenJev-VisionPython113GitHub stars
Example server that exposes Jev-style typed decisions and grounded span reads by seeding DiffusionGemma canvases and reading one denoise step per slot.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
mmastrac/djevPython
mmastrac/djevPythonFrom the Jev guides
Original write-ups on what this entry is for.
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
From the Jev community
Posts from builders shipping with Jev right now.