Jev Onnx

A small from-scratch alternative to Jev and Laya-style System 1 decision models: a roughly 22M-parameter shared encoder trained with per-example calibrated confidence labels in PyTorch and exported to ONNX for local, sub-second inference.

A Python decision engine described as calibrated and non-autoregressive, running in under 4 ms on CPU and powering dual-process agent reflexes, Antigravity IDE lifecycle hooks, an MCP server and OS automation; its system-one topics frame bounded, typed decisions, but the digest captured no README.

Run Laya locally on Bun with native ONNX inference and Hugging Face tokenizers: an HTTP server implementing TypeSafe's decision API plus an English/Persian playground with 16 demos, usable through the official @typesafe-ai/sdk with a custom baseURL.

Browser-only Laya demo suite: four Vite apps — a Traditional Chinese decision playground, Gomoku, Big Two and a 3D maze — run the Laya ONNX model on-device through Transformers.js, downloading fp16 weights from Hugging Face on first use.

Open, CPU-sized Jev-style decision model: give it a situation, a question and options, and Decima returns calibrated probabilities in roughly 20-50 ms per decision as int8 ONNX, with no GPU, no API key and nothing leaving the machine.

A Transformers.js ONNX conversion of open-jev-deberta-v3-large: fp16, fp32, q4 and q4f16 graphs from 0.35 GB that score typed questions in the browser through a single forward pass.

The largest of Bekko's v0 encoder models for Choice, Noul, and Score decisions: 395M parameters built on an Ettin reranker, with shared-prefix encoding and a 1.43 GB browser ONNX export.

Apache-2.0 ModernBERT fine-tune that answers noul, choice, score and multi-label questions with a probability for every option. It serves Jev's POST /v1/systemone locally so TypeSafe's Python SDK runs unchanged, reporting 0.026 expected calibration error on its test split.

Independent training toolkit and open weights for System One decision models in the same category as Jev: 17M, 68M, and 400M models answering Choice, Noul, Score, and reranking questions, with Python and ONNX inference and a documented generalization gap to Jev 1.13 on the project's S1MB benchmark.