Instructions to use Jibbscript/rptools-privacy-filter-multilingual-v2-mlx-8bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use Jibbscript/rptools-privacy-filter-multilingual-v2-mlx-8bit with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir rptools-privacy-filter-multilingual-v2-mlx-8bit Jibbscript/rptools-privacy-filter-multilingual-v2-mlx-8bit
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
File size: 1,306 Bytes
732be1f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 | ---
license: apache-2.0
library_name: mlx
pipeline_tag: token-classification
tags:
- mlx
- privacy
- pii-detection
- token-classification
- rptools
---
# RPTools Privacy Filter — Multilingual v2 (MLX, 8-bit)
On-device PII / privacy token classifier packaged for the RPTools / RedactFlow
Apple-Silicon MLX runtime (`rptools-mlx` manifest format, `openai-privacy-filter`
architecture family, BIOES-Viterbi decode). Ships the `rptools-mlx.json` manifest
the RPTools Swift loader expects.
**Provenance (honest):** the weights, tokenizer, config, and label map are a
byte-for-byte copy of
[`OpenMed/privacy-filter-multilingual-v2-mlx-8bit`](https://huggingface.co/OpenMed/privacy-filter-multilingual-v2-mlx-8bit)
(revision `df97c293841c59d253541960908e23b2d3a68212`). Only the MLX manifest is
re-emitted under the RPTools `rptools-mlx` format string (and `source_model_id`
rebranded); the load-bearing architecture identifier `openai-privacy-filter` is
preserved. Redistributed under the upstream Apache-2.0 license.
## Files
- `weights.safetensors` — 8-bit affine-quantized weights (group size 64)
- `tokenizer.json`, `tokenizer_config.json` — tiktoken-style tokenizer
- `config.json`, `id2label.json` — model config + 217-label map
- `rptools-mlx.json` — RPTools MLX artifact manifest
|