Fazmin/solus_v1_gliner2-privacy-filter-pii-q8

GLiNER2 Privacy Filter PII Multi (q8 ONNX) — Solus v1

The entity extractor behind Solus's PII tools: GLiNER2 Privacy Filter PII Multi, exported to ONNX and quantized to 8 bits.

It is a token classifier built on mdeberta-v3-base, not a generative model. It scores every candidate span in the text against a fixed label set and returns character offsets, so it cannot invent a finding that is not in the input. The label set is baked into the graph at export time: name, address, email, phone_num, id_num, url and username. The encoder's position budget is 512 subwords, so longer text has to be scanned in overlapping windows.

Specifications

Parameters	0.3B
Quantization	8-bit (MatMulNBits)
File size	525.14 MB
Minimum RAM	2.00 GB
Minimum VRAM	not required
Context length	512 tokens
SHA-256	261ee74758005ee265b999299e3d7091ddd541dceccdef414017a69beb4854a4 (model_q8.onnx)

Installs as a directory. Every file below is required:

File	Size
model_q8.onnx	1.28 MB
model_q8.onnx.data	508.58 MB
tokenizer.json	15.28 MB
entity_labels.json	239 B
onnx_export_metadata.json	1.54 KB

Quantization

Quantization performed at the Faculty of Engineering, McMaster University.

The conversion this build is derived from was produced by okasi, and the weights here are a byte-for-byte copy of that file — the SHA-256 above matches the upstream artifact.

Provenance

• Original model: fastino/gliner2-privacy-filter-PII-multi
• Upstream GGUF: okasi/gliner2-privacy-filter-pii-multi-onnx
• Mirrored for Solus, a desktop app for running language models entirely on your own machine.

Usage

import onnxruntime as ort

session = ort.InferenceSession("model_q8.onnx")

A token classifier, not a generative model. It scores candidate spans against a fixed label set baked into the graph at export time: name, address, email, phone_num, id_num, url, username. Preprocessing and decoding are described in onnx_export_metadata.json; model_q8.onnx.data must sit beside the graph.

License

Licensed Apache-2.0.