jevbooks

← All projects

jevmlx

bnsd55/jevmlx · Homepage

Parallel constrained decisions for MLX models on Apple Silicon.

What it is

jevmlx turns a schema of fields (booleans, enums, multi-selects) and a context string into a single batched forward pass on a local Apple Silicon model, returning schema-valid JSON with per-field probabilities.

How it uses Jev

For each schema field, jevmlx scores every allowed option from logits in one prefill, applies a restricted softmax, and assembles the JSON from the winners. The result is a Decision with per-field FieldResult (value, probability, alternatives, margin, ordinal telemetry).

Primitives:choicescore

Technique worth stealing

Restricted softmax over known options in one forward pass, avoiding token-by-token generation.

Try it

git clone https://github.com/bnsd55/jevmlx && cd jevmlx && ./setup.sh; jevmlx decide --preset support_triage --json

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.77
  • Shows a System One patternyes0.75
  • Handles uncertaintyno0.06
  • Measuredno0.05
  • Runnableyes0.97
  • Worth recommendingunclear0.44
  • Model replicayes0.82
  • Problem scopescore on a 0–2 scale1.12
  • About Jevyes0.62

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.