- Category
- Developer Tools
- Rank
- No. 1937Tools index
Previous survey · No. 1874 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Data Processing
- GitHub
- 5 stars
- Date
About
cbjev computes typed decisions (choice, score, or null) about a block of text or JSON state from a single encoder pass, letting multiple questions about the same document share one encoding. It's a self-hosted, faster, and better-calibrated successor to the author's earlier Laya model, and is wire-compatible with Jev.
At a glance
cbjev is a self-hosted decision model that answers several typed questions about the same text in one encoder pass. It is worth evaluating for repeated classifications when you can measure accuracy on your own examples.
Good fit
Teams building a local labeling or routing service that can fall back to a human when confidence is low.
Consider something else when
Use a different model when you need open-ended generation or require verified performance for a language absent from your test set.
Start here
Compare its choices on a small labeled fixture set with a simple baseline. Reorder the labels and record errors, confidence and latency on your hardware.
Cost and requirements
The repository documents Python 3.10 or newer, PyTorch and separately distributed weights. Self-hosting requires compatible compute.
Limits to know
The author measured latency on an RTX 4090. Its comparisons with hosted Jev use published figures from a different setup. Confidence needs calibration on your workload.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Gives engineers a faster, self-hosted way to extract structured choice/score decisions from text without paying per-question for a hosted classifier API or re-encoding the same document repeatedly.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.