GitHub · Research and data
featherless-ai/simple-jev
Seeing the output contract without a key.
What it does
Turns a compatible open model into something shaped like Jev: you send shared context plus a set of questions, and it reads the next-token logits for each question and assembles a JSON response of choices, rubric scores or truth/support judgements. No classifier head, no training run.
The repository's own description: “Turn any open model into a classifier/jev endpoint”
How it works
The server constructs the answer rather than asking the model to generate JSON, which is the same output contract by a different route. The repository ships `hf-server/`, an `eval/` directory, a `demos/jevpilot` demo and a `website/`.
The repository, by the numbers
Read from the GitHub API on 2026-09-22. Stars and the last push move daily — quote them with the date, the way we do.
What we checked
- the tree: common, eval, hf-server, demos/jevpilot, RFDT, website
- Apache-2.0 LICENSE
- README states the logits approach and that the model does not generate the JSON completion
- a public playground is linked from the README
What we did not check. We did not call the hosted playground, so we have not compared its scores with the hosted model's.