Awesome Jev-like models Jev Projects

Multilingual, non-autoregressive System 1 decision engine. Typed decisions over 100+ languages in a single forward pass — 33 ms — trained with reinforcement learning against strictly proper scoring rules (RLCD), with a router that picks the right checkpoint per request.

Open-model semantic-if engine for local Jev-style decisions.

Semantic ifs from open models, on a 3090 at home. Independent; not affiliated with Jev or TypeSafe.

Trainable family of small Qwen-based Jev-like decision models with typed primitives, datasets, evaluation tools, and local inference.

Open 4B decision model with prefill-only Choice, Noul, and Score inference over text or one image, plus native, vLLM, and SGLang runtimes.

Local 1B CPU decision model with Jev-compatible Noul responses; it returns P(yes) and explicitly rejects Choice and Score.

Local Qwen3.5 and Gemma 4 fine-tunes with a Jev-compatible API for typed Choice, Noul, and Score decisions in one forward pass.

Local Jev-compatible System One API that converts typed questions into DiffusionGemma classification prompts and probabilities.

The open-source System One decision model. Sub-15ms, non-autoregressive, local drop-in alternative to TypeSafe Jev.

Open-model Jev-style server that reads next-token logits to return typed choices, rubric scores, and truth judgments.

非自回归System 1决策引擎，专为结构化类型决策场景设计 DeepOpen Multilingual, non-autoregressive System 1 decision engine.

Run Laya, the open-source Jev-compatible System-1 decision model, from Node.js / TypeScript via ONNX Runtime.

Reasoning Jev-style 9B classifier with a diffusion drafter and compatible API for typed Noul, Choice, and Score decisions.

Multimodal Qwen3.5 decision models that answer multiple typed Choice, Score, and Noul questions with probabilities in one forward pass.

Jev-compatible API server built on open models and SGLang, with prefill-only inference and typed decision endpoints.

AgentJev-0.6B - a fast 'System One' decision model for AI Agents: feed it any unstructured state (diffs, traces, logs) and structured questions, get calibrated probability distributions back in one ~50ms forward pass. Zero output-token decoding.

Open probability decisions with Qwen3.5-2B, Qwen3.5-9B and Qwen3.8-27B.

The open, local take on Jev: typed decisions from an LLM, without generating a single token

Self-hosted Jev-compatible System One server backed by GLiNER-family models, with batching and typed question support.

Open bilingual mmBERT-based 144.3M decision model returning Choice, Noul, and 2–10 Score answers with probabilities through a separate local PyTorch runtime.

Open 4B and 35B-A3B video decision models that share a video prefix across questions and directly score temporal-grounding intervals and multiple-choice answers without text generation.

DiffusionGemma NVFP4 structured decisions on a DGX Spark: container recipe No repository-level license file was found during review.

Turn any LLM into a Jev-style decision model: typed decisions, real probabilities, no training. (continue updating).

Open, Jev-compatible System One decision server on DiffusionGemma.

Local multimodal decision models that answer Jev-compatible typed questions over records, text, and up to two images, with explicit abstention.

Same as jev-ultrafast but using Laya.

Adapter that turns local language models into Jev-style Choice, Score, and Noul decision engines with typed probabilities.

Contrastive language model for one-pass typed decisions and option probabilities.

OpenSource Jev.

Specialized CUDA inference backend for Open-Jev-27B with a drop-in Jev-compatible server, prefix-tree reuse, and checked benchmark artifacts.

Train small, insanely fast local classifiers from Jev-compatible examples. Run locally in your browser or Node.js.

Framework and fine-tuned embedding models that turn cosine similarities into Jev-style Choice, Score, and Noul probabilities.

Open-weight Jev alternative that returns typed decisions with probabilities in one forward pass.

Open semantic decision engine implementing Choice, Noul, and Score primitives with a TypeSafe comparison harness.

Open reproduction of TypeSafe Jev: a 150M typed decision engine (noul/choice/score in one non-autoregressive pass, calibrated confidence). 0.697 vs Jev's 0.727, 2.5x better calibrated, 4x faster, free. Trains on a Colab T4 in 30 min.

Type-safe language-model generation and structured decisions.

Solar Mini–based local model for Jev-style typed decisions.

Self-hosted API and web interface for Laya decision models with a Jev-compatible API shape.

Frozen open-model medical decision stack that scores each exam option, then uses confidence and conformal gates to auto-release or escalate answers.

Open visual decision model that encodes an image once and answers multiple structured questions.

Jev-style structured decisions on DiffusionGemma: the example server from vLLM PR 57250.

Training stack for a multimodal Jev-like System One model with vision.

Supersimple way to serve LLMs as a Jev-like endpoint.

Training and inference stack for calibrated Jev-style choice models with dynamic candidate sets.

Turning an LLM model into a Jev like System.

Unofficial PoC: a BERT-style encoder decision engine behind a typed-decision (noul / choice / score) HTTP API. Not affiliated with TypeSafe.

open-jev is a browser-focused TypeScript library for typed decisions: one piece of text (the state) plus any number of typed questions go in, and one forward pass returns a calibrated probability distribution per question. Nothing is generated, so an answer is always one of the options you provided.

Open Thai-English System One model with Choice inference, calibration, evaluation scripts, and a Jev comparison suite.

Open model for typed, calibrated decisions in one forward pass.

Open probabilistic Choice, Noul, and Score API powered by constrained open-model log probabilities.

Local Qwen3.5-0.8B decision model using llama.cpp label probabilities for Noul, Choice, and Score, with a Jev-compatible HTTP endpoint.

Open 4B English/Korean decision model with Jev-compatible Choice, Score, and Noul probabilities plus BF16 and GGUF weights.

Rust System One–compatible API for open decision models such as Laya.

Serve typed-decision (System 1) models — Laya or your own — on NVIDIA GPUs or Apple Silicon, with a Jev-compatible API and coding-agent integrations.

Local Jev-compatible proxy that learns an application’s typed decisions and serves them with a Laya head.

Super fast Jev like server, model agnostic, working with any OpenAI compatible endpoint.

Local Jev-compatible typed decisions from unstructured state in one forward pass.

Typed JSON inference with DiffusionGemma, with Every and Jev benchmark results.

Run SemIf (Jev-style semantic-if decisions) on a CPU — no GPU. Reads typed option probabilities straight from an open model in one forward pass, plus a web UI.

Local decision-model runner with a Jev-compatible System One API.

Open decision models for finance with a Jev-compatible typed API and evaluation runners.

Open-weight Jev-inspired decision model fine-tuned from Qwen3.5 4B.

Run Jev-style typed decisions locally on your Mac with low RAM usage and fast responses.

High-performance System One model with an embeddable C runtime and WebAssembly support.

Open Jev-style model and training stack for typed decisions and option probabilities.

Ask typed questions about text or JSON and inspect model-derived probability distributions in a browser. One container runs the model, the original Decision Lab explorer, and a TypeSafe-shaped API.

an open-source, from-first-principles reconstruction of the ideas behind TypeSafe AI's Jev, written in pytorch.

Typed decisions (choice / score / yes-no) from local Qwen models on Apple Silicon. Probabilities come straight from the logits, no text generation. TypeSafe-compatible HTTP API, runs on MLX.

Local Jevfire-style candidate scoring with ternary models and a System One API; probabilities are candidate-relative.

Qwen3.5-based decision models accepting text and images through a Jev-compatible server, with typed answers from label-token probabilities; weights are non-commercial.

Jev-shaped classification wrapper over OpenAI-compatible clients that returns probabilities and confidence.

Local 0.8B Jev-style model and compatible server for Choice, Score, and Noul decisions, with agent skills and a Claude Code permission guard.

Open multimodal decision model that returns candidate probabilities for text and screenshots through a Jev-style Choice, Score, and Noul endpoint.

Local Jev alternative serving Noul, Choice, Score, and multi-label probabilities through a System One-compatible endpoint.

Open local System One implementation for NVIDIA DGX Spark, with a Qwen3-based model, Jev-compatible API, architecture notes, and evaluation tooling.

Open 2B and 4B multilingual decision models focused on Korean, serving Jev-compatible Choice, Score, and Noul probabilities through Kev.

Local Jev-like System One decision layer that returns typed choices and calibrated confidence without a hosted API.

OpenJev: an independent Jev-inspired System One decision API based on TypeSafe.ai concepts. Choice, score and noul primitives, local mock server, Python and TypeScript SDKs. Real inference planned; not affiliated with TypeSafe AI. No repository-level license file was found during review.

Local Jev-like typed-decision runtime using ONNX and quantized models for offline CPU inference.

Open multimodal System One model in four sizes that answers typed questions over screenshots, images, video, or text in one forward pass.

Local 596M typed-decision model with its own weights, Choice, Score, and Noul support, and a Jev-compatible System One endpoint.

Local Jev-compatible evaluation server: POST /v1/systemone with typed noul/choice/score, open weights, no waitlist No repository-level license file was found during review.

Local decision stack that fuses four readings from two frozen open models, then uses agreement and calibration gates to answer or abstain.

Independent DiffusionGemma and SGLang typed-decision server for text and images, with a drawing playground and public evaluation artifacts.

Independent multilingual decision model with calibrated answers, abstention, and a Jev-compatible API.