AI Attributes™Models
AI Attributes curates and develops open models that read attributes of text. Every model is published with open weights on Hugging Face, so you can download it and run it yourself, or call it through its product's API.
There are 6 today: for each attribute, a small RoBERTa model and a larger Gemma model, fine-tuned for the same task so they can be compared on the same text.
| Reads | Model | Base | Labels | Licence |
|---|---|---|---|---|
| Emotions | open-emotions-roberta-125m-fp32 | RoBERTa | 28 | MIT |
| Emotions | open-emotions-gemma4-e2b-bf16-lora | Gemma | 28 | Apache-2.0 (adapter) |
| Moral framing | open-morals-roberta-125m-fp32 | RoBERTa | 8 | MIT |
| Moral framing | open-morals-gemma4-e2b-bf16-lora | Gemma | 8 | Apache-2.0 (adapter) |
| Sensations | open-sensations-roberta-125m-fp32 | RoBERTa | 6 | MIT |
| Sensations | open-sensations-gemma4-e2b-bf16-lora | Gemma | 6 | Apache-2.0 (adapter) |
Where each one lives
- OpenEmotions™: its models, what they score, and its API.
- OpenMorals™: its models, what they score, and its API.
- OpenSensations™: its models, what they score, and its API.
Using them
- Through the API. Each product has one: send blocks of text, get a score for every label. Any platform can connect with a URL and a key, as AI Observability Studio and Am I Conscious? do.
- On your own machines. Download the weights from Hugging Face. Nothing you score leaves your systems.
Either way, a result names the model and the exact revision that produced it.
How well they read
Each model's card on Hugging Face reports its results on a held-out test set, with the training settings and the data it used. A score is a model's reading of text, and the method and ethics say what that does and doesn't mean.
More to come
Emotions, moral framing and sensations are the first three attributes. Others will follow, and others can contribute a model.