File size: 1,902 Bytes
794b114 a2a5f1d 794b114 a2a5f1d 794b114 a2a5f1d 48eda4c 794b114 48eda4c a2a5f1d 5ff9346 f984f65 5ff9346 a2a5f1d 5ff9346 a2a5f1d 5ff9346 f984f65 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 | ---
title: Decida
emoji: 📊
colorFrom: indigo
colorTo: gray
sdk: gradio
sdk_version: 5.9.1
app_file: app.py
pinned: false
license: apache-2.0
short_description: "DecidaBERT-large: v1.1 vs v2, typed decisions in one pass"
---
# Decida — DecidaBERT-large: v1.1 vs v2
A free ZeroGPU demo that runs [DecidaBERT-large-v1.1](https://huggingface.co/helmo/DecidaBERT-large-v1.1) and
[DecidaBERT-large-v2](https://huggingface.co/helmo/DecidaBERT-large-v2) side by side: send a **state** and typed
**questions** (`choice`, `score`, `noul`), get a probability distribution per question from each model, one forward
pass each, no generated text.
- **Benchmark cases:** any of the 400 cases of the Typed Decisions **test** split (never used for training), with the
benchmark's gold answers. Each model is marked ✓ or ✗ per question, and the gold distribution is shown beside them.
- **Hand-written demos:** a long incident log, a pull request with twelve questions, a phishing email, an invoice
that does not match its order, and more. Every question carries the demo author's reading of the right answer
(a judgment, not a benchmark label), so you can see where the models match it, and where they agree or disagree
with each other.
- **Your own:** write any state (text or JSON) and questions, or build them with the question form; add `"expected"`
to a question to have both models marked against it.
The **Benchmark** tab has both models' scores on the Typed Decisions test split, overall, per question type and per
workflow. The two are level overall and differ by workflow.
This Space's code is independent of the main project: `decidabert_v2/` is the inference code published with v2, and
it loads both models (same architecture). The local runtime with a REST API, MCP tools and a testbench is at
**https://github.com/heldernoid/decida**.
Not affiliated with any other model author.
|