---
license: apache-2.0
library_name: pytorch
tags:
- decision-model
- multimodal
---

# WeMM-JEV

WeMM-JEV is a 27B multimodal decision model based on Qwen3.8-27B. 
It supports choice, noul (yes/no) and score questions over text and multiple images, returning probabilities in a single forward pass per question.

- **Choice:** select among supplied options.
- **Noul:** estimate the probability of a yes/no proposition.
- **Score:** return a probability distribution over ordered levels.

## Public Decision Index 0.3

| Panel | Public index |
| --- | ---: |
| Text | 69.37 |
| Vision | 77.97 |

[Detailed evaluation results](https://huggingface.co/datasets/Rover00/WeMM-JEV-Results)

## Usage

Install the dependencies in [inference/README.md](inference/README.md), then run
from your local model directory:

```bash
cd /path/to/WeMM-JEV
python - <<'PY'
from inference import D3Engine

engine = D3Engine(model=".", device="cuda:0")
response, _ = engine(state="The light is red.", questions={
    "color": {"type": "choice", "instructions": "What color is the light?",
              "criteria": {"A": "Red", "B": "Blue"}}
})
print(response)
engine.close()
PY
```

See [text and image examples](inference/README.md#python),
the [HTTP server](inference/README.md#server) and
[Decision Index integration](inference/README.md#decision-index).