jevbooks

← All projects

edgejev

yzfly/edgejev · Homepage

Offline typed decisions on CPU: 15.6 ms per question, no torch at runtime.

Infra / SDKs / Integrations0.53Runner-up: Classification & RoutingQuestionnaireCheck-up
Automated README summary

Automatically summarized from the README on 1 Nov 2026; not a run by jevbooks. Run instructions and author-reported outcomes still need verification.

What it is

EdgeJev unifies open-source Jev/System One reproductions (laya, kev, NanoJev, PlayJev) into a convert→quantize→deploy pipeline: export ONNX, INT8 quantize, and serve a local protocol-compatible API on ordinary CPUs, fully offline.

How it uses Jev

Jev makes typed decisions (choice, score, noul) on a given state. EdgeJev runs these decisions locally via ONNX Runtime, returning calibrated probabilities. The Agent.system_one method takes a state and a dict of questions, returning answers with probabilities, confidence, and choices.

Primitives:choicescorenoul

Technique worth stealing

ONNX graph optimization plus INT8 quantization with AVX512-VNNI integer multiply-add, cutting latency and model size while staying bit-identical to upstream in fp32.

Try it

uv tool install "edgejev[build]" && edgejev build --backend laya --out ./jev-int8

View on GitHub

Start using Jev: API key and your first request →

Model judgments from README excerpts

Project signals use repository metadata, an existing summary and the first 700 README characters, not the full project. “Not found” means not detected in that material; it does not establish absence. These are the original judgments with recorded models and dates, not source audits or run results.

Models & evaluation dates

Inclusion gate
jev-1.14.0
Project signals
jev-1.14.0
Category
jev-1.14.0
Pattern tags
jev-1.14.0

Evidence

These signals use repository metadata, an existing summary and at most 700 cleaned README characters. ≥ 0.60 means reported in that excerpt; ≤ 0.40 means not found there. The middle band is inconclusive. None is a source-code audit or a test result. The runnable-tool question excludes research experiments; the recommendation question asks for a learning shortlist, not a quality verdict.

  • Jev is core to the projectFound in material0.95
  • Shows a System One patternnot found in the excerpt0.04
  • Describes handling uncertaintynot found in the excerpt0.02
  • Reports measured numbersFound in material0.93
  • Is a runnable toolFound in material0.96
  • Top-10 learning shortlist signalunclear0.51
  • Is a replica of the modelFound in material0.92
  • Solves a common problem (0–2)score on a 0–2 scale1.17 / 2
  • About JevFound in material0.99
Signals
Jev is core to the project0.95Shows a System One pattern0.04Describes handling uncertainty0.02Reports measured numbers0.93Is a runnable tool0.96Top-10 learning shortlist signal0.51Is a replica of the model0.9200.51Solves a common problem (0–2)1.17 / 2012

no-judgment band (0.40–0.60)

Design patterns
Questionnaire0.92Check-up0.86Prefetch0.2100.51

no-judgment band (0.40–0.60)

The Jev model and date for each judgment stage are listed above. Card written by DeepSeek V4.1 Flash from the README on 1 Nov 2026.