README.md

---
title: Jevásio
emoji: ⚖️
colorFrom: green
colorTo: red
sdk: gradio
sdk_version: 6.28.0
python_version: "3.12"
app_file: app.py
short_description: Typed decisions for European Portuguese text
startup_duration_timeout: 1h
license: cc-by-nc-sa-4.0
models:
  - TigreGotico/jevasio-0.8b
  - TigreGotico/jevasio-0.8b-v2
  - TigreGotico/jevasio-4b
  - TigreGotico/jevasio-4b-v2
  - Qwen/Qwen3.5-0.8B-Base
  - Qwen/Qwen3.5-4B-Base
datasets:
  - TigreGotico/jevasio-decisions
  - TigreGotico/jevasio-decisions-v2
tags:
  - decision-model
  - european-portuguese
  - pt-pt
---

# Jevásio

Jevásio is a decision model for European Portuguese (pt-PT). It reads a state, which is a text or a JSON object, and a
set of typed questions about it. It returns a probability for each option of each question in one forward pass. It
does not generate text.

This Space runs four models on ZeroGPU, in bf16. Pick one with the model selector. The 4B models are more accurate, and
the 0.8B models are smaller.

- [Jevásio-4B-V2](https://huggingface.co/TigreGotico/jevasio-4b-v2) and
  [Jevásio-0.8B-V2](https://huggingface.co/TigreGotico/jevasio-0.8b-v2), trained further on more tasks: the language
  variety of a text, voice commands, news sections, legal areas, sentence relations, offensive language, puns, and
  translations or summaries with one fact changed. The examples marked (V2) are from those tasks.
- [Jevásio-4B](https://huggingface.co/TigreGotico/jevasio-4b) and
  [Jevásio-0.8B](https://huggingface.co/TigreGotico/jevasio-0.8b), the first versions.

Jevásio-4B and Jevásio-0.8B are [Kev](https://github.com/jaredpalmer/kev) fine-tuned on the
[TigreGotico/jevasio-decisions](https://huggingface.co/datasets/TigreGotico/jevasio-decisions) dataset. The V2 models
continue from them on
[TigreGotico/jevasio-decisions-v2](https://huggingface.co/datasets/TigreGotico/jevasio-decisions-v2).

## Request shape

A request follows Kev's System One format (`POST /v1/systemone`). Keys are English: the state's field names, the
question ids and the option names. Values are Portuguese.

```json
{
  "state": {"request": "Fui cobrado duas vezes pela encomenda 1182."},
  "questions": {
    "team": {"type": "choice", "instructions": "Que equipa deve tratar deste pedido?",
             "criteria": {"billing": "Cobranças e reembolsos", "shipping": "Entregas", "returns": "Trocas"}},
    "urgent": {"type": "noul", "instructions": "Este pedido precisa de resposta hoje?"},
    "frustration": {"type": "score", "instructions": "Quão frustrado está o cliente?",
                    "criteria": ["Calmo", "Incomodado", "Muito zangado"]}
  }
}
```

There are three question types. A `choice` question picks one of its named options, and each option can have a
description. A `noul` question is yes or no, and the answer is p(yes). A `score` question picks a level on an ordered
list, and the answer is the expected level.

The **Perguntas** tab builds a request from a form. The **Avançado** tab takes a raw JSON request and shows the raw
JSON answer. The same two functions are on the Gradio API as `/decide` and `/systemone`.

## How it runs

The Space installs Kev from GitHub at commit `1f04cfa28e6a708485b28a9585010b9e436b776e` and loads the model with
Kev's own loader, as `python -m kev.serve --run TigreGotico/jevasio-0.8b` does. The probabilities use the calibration
temperature stored in the model's `head.pt`.

## Licence and limits

The models and the dataset are under CC BY-NC-SA 4.0, for non-commercial research only. The model cards list the
measured accuracy, the calibration error and the known limits. Do not use the model for decisions with legal, medical,
financial or employment consequences without human review.