[Breaking] Jev runs on llama.cpp.

@TORA276990 TORA · Sep 20, 2026 · case #993780 · listed via Jev Hub

Still from the clip posted with this case; watch it in the original post on X.

Implemented a Jev-compatible API in llama-server so local GGUF models return scored judgments, hitting 0.891 accuracy at 154ms per question.

The original post, translated

[Breaking] Jev runs on llama.cpp. Developer kishida has implemented a Jev-compatible API for TypeSafe's judgment-only AI "Jev" into llama-server. Any GGUF on hand becomes a judgment engine.
- 0.891 accuracy with Gemma 4 12B, 154ms per question
- No training required, images OK too
- Returns with probabilities, zero generation
Local Jev is now within reach.

Show the original post in its source language ▾

【速報】llama.cppでJevが動く TypeSafeの判断専用AI「Jev」互換APIを開発者kishida氏がllama-serverに実装。手元のGGUFが何でも判定機になる・Gemma 4 12Bで正答率0.891、1問154ms ・学習不要、画像もOK ・確率付きで返す、生成ゼロローカルJev、もう手が届く

The post above is a machine translation from ja; the untranslated text is in the fold-out.

Engagement when collected

Likes1

Bookmarks0

Reposts0

Replies0

Quotes0

Numbers are a snapshot taken from X when the case was added to the library (schema v1, collected 2026-09-19); they will not match today.

Where this case fits

Filed under coding & developer tools. In the pattern Jev is built for, the model answers a bounded question per step — and ordinary code acts on the answer, because the answer is already a value rather than a paragraph. Other posts in the same family are on the coding & developer tools page.

🛠️ Coding & developer tools