| --- |
| 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. |
| ``` |
|
|
