jeva.cpp
github.com/pragmatwice/jeva.cpp- Category
- Developer Tools
- Rank
- No. 2100Tools index
Previous survey · No. 2004 ·
- Type
- TOOL
- Use case
- Models: Train & Run
- Interfaces
- API
- GitHub
- 4 stars
- Latest release
- b11230
- Date
About
jeva.cpp is a fork of llama.cpp that adds a JEV-compatible decision API, so any large language model already supported by llama.cpp can be run as a structured decision model returning typed judgments rather than free-form text.
At a glance
jeva.cpp adds a structured decision API to a llama.cpp server so a local model can score or choose options from its next-token outputs. It is worth evaluating if you want typed decisions alongside your existing local inference workflow.
Good fit
Developers who already serve a compatible model and want to benchmark small classification tasks.
Consider something else when
Keep the ordinary generation endpoint when you need free-form answers or your model cannot expose vocabulary logits.
Start here
Use an approved local model and a small set of labeled choices. Compare accuracy and latency with your existing classifier and repeat after reordering the options.
Cost and requirements
MIT-licensed fork requiring a compatible llama.cpp backend and model. Decision endpoints require next-token vocabulary logits.
Limits to know
An API-compatible response does not establish comparable confidence or accuracy across models. Validate both on your own inputs.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Lets developers run calibrated decision models locally on any llama.cpp-supported LLM, avoiding cloud API calls for simple classification judgments embedded in agent workflows.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.