jev-test
clduab11/jev-test
Pre-registered benchmark: can a 2B local model (Gemma 4 E2B) answer web questions without making things up when a decision model (TypeSafe Jev) makes every call? SearXNG for search, MemPalace for verbatim memory, seven arms including open local judges. Spec and thresholds fixed before any run.
- Stars
- 1
- Forks
- 0
Why it is listed
jev-test is a research harness using TypeSafe Jev as a separate judge for search, evidence selection, and sentence support; its README also reports the judged pipeline missed its baseline.