---
license: apache-2.0
language:
- en
library_name: transformers
pipeline_tag: text-classification
base_model: distilroberta-base
datasets:
- aitrainingdatabroker/synthetic-contract-clauses-by-type-and-risk
tags:
- contracts
- legal-nlp
- clause-classification
- data-licensing
- synthetic
---

# Contract Clause Type Classifier

Reads one data license clause and names its type out of eight: exclusivity, term and renewal, payment, confidentiality, IP and ownership, indemnity, audit and deletion, or termination.

Every name, number and sentence it was trained on is made up. Nothing comes from a real person, customer or business.

Related tool: [contract-clause-watchlist](https://huggingface.co/spaces/aitrainingdatabroker/contract-clause-watchlist) (a Space that runs in your browser).

## Why it exists

A data license is easier to review when each clause is filed under its type. This model files clauses so a person can compare, say, every indemnity across offers.

## Use it

```python
from transformers import pipeline

classify = pipeline("text-classification", model="aitrainingdatabroker/contract-clause-type-classifier", top_k=3)
print(classify("The Licensee may keep the Dataset after the term and need not delete it."))
```

## How it was trained

Fine-tuned from `distilroberta-base` for 6 epochs on one RTX 3070. Training data: 376 rows from [synthetic-contract-clauses-by-type-and-risk](https://huggingface.co/datasets/aitrainingdatabroker/synthetic-contract-clauses-by-type-and-risk), all invented, with test rows built from templates the model never saw in training. The scripts that built the data and trained the model are in `source/`; the full scores are in `scores.json`.

## Test scores

On 186 held-out rows: accuracy 0.726, macro F1 0.711.

| Label | F1 | Count |
|---|---|---|
| `audit_and_deletion` | 0.533 | 24 |
| `confidentiality` | 1.0 | 22 |
| `exclusivity` | 0.683 | 23 |
| `indemnity` | 0.96 | 24 |
| `ip_and_ownership` | 0.9 | 22 |
| `payment` | 0.657 | 24 |
| `term_and_renewal` | 0.957 | 24 |
| `termination` | 0.0 | 23 |

## Limits

- Trained only on invented text from templates, with test sentences built from templates it never saw. Real text varies far more.
- Real clauses are longer and mix topics; a clause with two ideas gets one label.
- This model is not legal advice and not a certification. A person should review its output before anything is shared or decided, and your counsel decides what a rule requires.

## Read more

- [How you get paid when you license data](https://aitrainingdatabroker.com/sell/how-you-get-paid/?utm_source=huggingface&utm_medium=model&utm_campaign=contract-clause-type-classifier)
- [Glossary of data licensing terms](https://aitrainingdatabroker.com/glossary/?utm_source=huggingface&utm_medium=model&utm_campaign=contract-clause-type-classifier)
- [Who is liable when data is licensed to AI](https://aitrainingdatabroker.com/sell/liability/?utm_source=huggingface&utm_medium=model&utm_campaign=contract-clause-type-classifier)

## License and credit

Apache 2.0. Please credit **AI Training Data Broker** and link to https://aitrainingdatabroker.com/ when you share or build on it.

## About

AI Training Data Broker is a broker and technical advisor for companies that sell or license their data to AI labs. It guides the technical side, including export, format, de-identification and compliance, and publishes free guides, checklists and tools.