patronus-studio/gliner_small-v2.5-edge

GLiNER small v2.5 — Edge (ONNX, quantized)

An edge-optimized ONNX build of GLiNER small v2.5, a zero-shot named-entity recognition model that extracts arbitrary user-supplied entity labels.

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	gliner-community/gliner_small-v2.5 — Apache License 2.0
Backbone	microsoft/deberta-v3-small — 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 PyTorch model to ONNX.
• Applied post-training dynamic quantization: INT8 weights for MatMul operations and INT4 for the token-embedding matrix, 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
model_int4_embeddings_int8.onnx	Quantized ONNX graph (INT8 weights, INT4 embeddings)
spm.model, tokenizer.json, tokenizer_config.json, special_tokens_map.json	Tokenizer (unchanged from upstream)
gliner_config.json, gliner_onnx_config.json	GLiNER and ONNX runtime configuration
quantization_manifest.json	Records the quantization applied

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.