---
license: apache-2.0
base_model:
- fastino/gliner2-multi-v1
pipeline_tag: token-classification
tags:
- gliner
- ner
- zero-shot
- pii-detection
- privacy
- multilingual
- onnx
- quantized
- edge
language:
- de
- en
---

# GLiNER2 multi — Edge (ONNX/ORT, quantized)

An edge-optimized ONNX/ORT build of **GLiNER2 multi**, a multilingual zero-shot information-extraction model.

Packaged for on-device inference; used as the model behind the `dynamic-pii` pipeline of **Patronus Ark**, Patronus' open-source AI-security scanning library. Ark is not publicly released yet — a repository link will be added at launch. The model itself is the work of the upstream authors — see [Attribution](#attribution).

## Attribution

| | |
| --- | --- |
| **Upstream model** | [`fastino/gliner2-multi-v1`](https://huggingface.co/fastino/gliner2-multi-v1) — Apache License 2.0 |
| **Backbone** | [`microsoft/mdeberta-v3-base`](https://huggingface.co/microsoft/mdeberta-v3-base) — MIT License |

The upstream Apache-2.0 license and its notices are retained. The MIT terms continue to apply
to the portions originating from the backbone model.

## Changes made to the original work

In accordance with Section 4(b) of the Apache License 2.0, the modifications relative to the
upstream model are:

- Exported the upstream model components to **ONNX** and to the **ORT** runtime format.
- Applied **post-training dynamic quantization** (INT8 weights; INT4 embeddings in the encoder) to reduce disk and resident memory for CPU/edge inference.
- No retraining, fine-tuning, or change to the model architecture or learned behavior beyond the numerical effects of quantization.

## Contents

| File | Purpose |
| --- | --- |
| `encoder_int8_int4.onnx` / `.ort` | Quantized encoder (INT8 weights, INT4 embeddings) |
| `classifier_int8`, `span_rep_int8`, `count_pred_int8`, `count_lstm_int8` | Quantized task heads (ONNX and ORT) |
| `spm.model`, `tokenizer.json`, `added_tokens.json` | Tokenizer (unchanged from upstream) |
| `config.json`, `gliner2_config.json`, `edge_manifest.json` | Model and edge-build configuration |
| `required_operators.config` | Operator set required by the ORT build |

## License

This work is distributed under the **[Apache License 2.0](https://www.apache.org/licenses/LICENSE-2.0)**,
the same license as the upstream model. A full copy of the license is included as `LICENSE` in
this repository.

```
Copyright the original GLiNER authors and contributors.
Modifications (ONNX export and quantization) Copyright Casdo Labs GmbH.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
```