instruct-jev
Instruction dataset of 119 rows compiled from TypeSafe's public Jev documentation, split by Choice, Noul and Score.
This repository is mostly documentation or write-up rather than runnable code.
Overview
INSTRUCT_JEV is a small instruction corpus compiled by DeckerGUI from TypeSafe AI's public Jev documentation. It contains 119 instruction rows (47 choice, 51 noul, 21 score) drawn from 114 unique documents, and is published on Hugging Face as ctaxnagomi/INSTRUCT_JEV with per-primitive JSONL subsets, metadata and statistics.
It is intended for people building or fine-tuning Jev-style decision models, or tools that need examples of how the primitives are phrased. A standard-library Python builder re-mirrors the cleaned source corpus and regenerates the dataset idempotently. The README credits TypeSafe AI as owner of the concepts and documentation, used under the MIT-licensed public docs. It is a data project rather than an application, and the corpus is small. Contributions are welcome via pull request.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Other | Instruction rows illustrating Choice, Noul and Score questions | choice · noul · score |
How it works
- Capture TypeSafe documentation sources
- Run the builder to extract instruction rows
- Publish train.jsonl and per-primitive subsets to Hugging Face
- Load with datasets.load_dataset
Strengths
- Ready-made examples of the three primitives
- Reproducible build from source documents
- Available directly through Hugging Face datasets
Caveats
- Only 119 rows; too small for training on its own
- Derived from documentation, not real decision traffic
Getting started
`from datasets import load_dataset; ds = load_dataset("ctaxnagomi/INSTRUCT_JEV")`.At a glance
- Type
- Project
- Status
- research
- Language
- Python
- Runtime
- dataset
- Has code
- yes
Use cases and integrations
How curated lists describe it
INSTRUCT_JEV - TypeSafe AI Jev / System One instruction corpus…
wh000wh000/awesome-jev-liveINSTRUCT_JEV - TypeSafe AI Jev / System One instruction corpus (choice/noul/score), compiled by DeckerGUI. 119 rows. Mirrored on HuggingFace.
hellogumbo/awesome-jev
Frequently asked questions
What is in the dataset?
119 instruction rows built from Jev docs: 47 choice, 51 noul and 21 score.
Can I train a model on it?
It is small, so it is better suited to evaluation, prompting examples or seeding a larger set.
Who owns the content?
The README credits TypeSafe AI as owner of the concepts and source documentation.
Related projects
33-chapter Chinese tutorial on agentic RL and GRPO, including a chapter on when Jev-style external decision models beat training with RL.
ProjectresearchPythonKorean study notes and TypeScript mock-ups of Jev concepts, with a local Laya versus GPT-5 comparison and a calibration toolkit.
ProjectresearchTypeScriptFrom-first-principles PyTorch reconstruction of a Jev-like architecture with typed noul, choice and score heads; random weights only.
ProjectresearchPythonLearning project that builds choose-an-answer decision models from scratch and from Qwen, with downloadable 0.8B and 4B releases.
LibraryresearchPythonMinimal training project for a Jev-style decision model: Qwen3-0.6B with a pointer head that outputs probabilities over options.
ModelresearchPythonPreprint survey of Jev-style typed decision models: their research roots, adaptation costs and what current evidence supports.
Modelresearch