patronus-studio/gliner2-multi-edge

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

Upstream model	fastino/gliner2-multi-v1 — Apache License 2.0
Backbone	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, 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.