privacy-filter-onnx

8-bit ONNX exports of two named-entity recognition models, for finding people, places and organizations in text on the user's own computer.

Folder	Original model	Revision	License
he/	dicta-il/dictabert-large-ner by DICTA	ac70130828d5a6811c2dc451e2255424c96f0eb0	CC BY 4.0
en/	dslim/bert-base-NER by David S. Lim	d1a3e8f13f8c3566299d95fcfc9a8d2382a9affc	MIT

Each folder holds:

• model_int8.onnx: the float32 model exported with torch.onnx.export (opset 17), then quantized with onnxruntime.quantization.quantize_dynamic (QInt8 weights). Inputs input_ids, attention_mask, token_type_ids; output logits.
• tokenizer.json, tokenizer_config.json: the original tokenizer.
• labels.json: the original model, its revision, and id2label.

The weights are the original authors' work, changed only by the export and the quantization. Please cite and credit them as their model cards ask.