Kev
⚡
37
tiny Jev-like decision models you can train yourself
Kev
Jev-inspired decision models: typed questions in, calibrated probabilities out, one forward pass. Named by base model size.
Note Try Kev-4B and Kev-0.8B in the browser (ZeroGPU). Same encoder and API code as kev.serve.
jaredpalmer/kev-27b
Text Classification • Updated • 1Note Kev-27B: Qwen3.8-27B (post-trained), bf16, H200 / H100 80 GB
-
jaredpalmer/kev-9b
Text Classification • Updated • 1.46k • 40 -
jaredpalmer/kev-4b
Text Classification • Updated • 6.09k • 43 -
jaredpalmer/kev-0.8b
Text Classification • Updated • 4.31k • 18
jaredpalmer/kev-0.5b
Text Classification • Updated • 1.32k • 46Note v0.1 prototype · Qwen2.5-0.5B · acc 0.799 / ECE 0.065 (in-distribution). Use Kev-0.8B+ for production work