AgentJev: a small candidate-head model for agent decisions

A Qwen3-0.6B backbone with its language-model head removed scores caller-supplied candidates through a permutation-equivariant head. The project releases training, serving and prefix-sharing code for Boolean, Choice and Score decisions, including a coding-completion checkpoint.

Notes

The README's immutable BF16 v1 release reports 1,585/2,000 correct teacher-labelled answers (79.25%) from 400 Typed Decisions cases, with chosen-option ECE 0.0448; its current default coding-completion checkpoint is a separate release.

source-backed — Public repository, docs or live artifact. About this label

Destinations

Availability & demo checks

Report a problem →

Also linked

• Immutable AgentJev BF16 v1 weights ↗
• Current AgentJev coding-completion checkpoint ↗

Filed under

• #local
• #training
• #agents
• #decision-making

In collections

• Agent harnesses →
• Local & replica models →