Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

Training-free inference techniques with decision models

Ways to get decision-model behaviour from ordinary LLMs, mostly by reading option logits in one forward pass instead of generating text.

128 public projects · page 5 of 5

  • JEVfire

    Jev-inspired technique scoring single-token labels with an LLM's head via vLLM to fill typed fields, with a fast Mario demo.

    BenchmarkresearchPython
  • open-jev (JoshuaSP)

    Experimental harness for typed JSON decisions from DiffusionGemma by picking the likeliest allowed tokens from final logits.

    BenchmarkresearchPython
  • SGLang

    SGLang's native /v1/decisions and /v1/systemone endpoints that read option probabilities from a validated chat model's next-token scores.

    ServicePython
  • Build your own decision model

    Tutorial contrasting one-pass option probabilities from a decision model with an LLM that needed 11 passes to emit the same JSON.

    Articlediscussion
  • TypeSafeのJevを正しく驚く、それってLLMでできませんか？

    Japanese article testing whether LLMs can replicate Jev, via a logit shortcut on Gemma and the Mario harness.

    Articlediscussion
  • It is the inference technique, not the training

    X thread arguing Jev's speed comes from parallel decoding, so a Jev-like API can sit over any open-weight model.

    Discussiondiscussion
  • Jev-compatible public API

    Social post about a public Jev-shaped API served by an open Qwen3.6-35B-A3B model so anyone can try typed decisions.

    Discussiondiscussion
  • Show HN: jev-bisect – Find answers with Jev via bisection search

    Show HN thread for jev-bisect, a pip package that finds answers with Jev by bisection search over options or ranges.

    DiscussiondiscussionPython