Category	What it detects	Examples
`company_name`	Supplier names, customer account names, corporate entities	`Meridian Logistics Ltd`, `Apex Manufacturing PLC`, `TechStart Solutions Inc`
`price`	Monetary values with currency symbols or codes	`GBP 4,250.00`, `$12,500`, `EUR 5,000`, `2,500 GBP`
`id_number`	Purchase orders, invoice numbers, case IDs, internal references	`PO-00442`, `INV/2024/00567`, `CASE-20240315-001`, `ORD-78542`

The model and Excel script run locally. Your Excel data is read and processed on your own machine.

What This Repo Includes

This model repo contains:

model.safetensors and config.json - the finetuned OPF checkpoint.
clean_excel.py - a helper script for cleaning Excel workbooks.
requirements.txt - Python packages needed by the helper script.
dummy_procurement_data.xlsx - a sample workbook for testing the setup.
label_space.json - the full label list used by the model.

What The Excel Script Does

clean_excel.py takes an input .xlsx file and creates a new cleaned workbook.

It:

Loads the workbook.
Detects sensitive values in selected text columns.
Creates synthetic replacements using Faker.
Adds new <column>_clean columns containing the synthetic text.
Adds a PII_Audit_Log sheet showing what was detected and replaced.
Leaves the original workbook untouched.

Repeated values are consistent within a single run. For example, if Meridian Logistics Ltd appears ten times in one workbook, it will be replaced with the same synthetic company each time in that run.

Requirements

You need:

Python 3.10 or newer.
Internet access for the first install and first model download.
Enough disk space for the model checkpoint.

Use CPU unless you already have a working CUDA setup.

Fresh Install On Windows

Run these commands in PowerShell:

powershell

mkdir privacy-filter-demo
cd privacy-filter-demo

py -3.12 -m venv .venv
.\.venv\Scripts\Activate.ps1

python -m pip install -U pip huggingface_hub

hf download galexdav/privacy-filter-finetuned `
  --include clean_excel.py `
  --include requirements.txt `
  --include dummy_procurement_data.xlsx `
  --include label_space.json `
  --local-dir .

python -m pip install -r requirements.txt

If py -3.12 does not work, use your installed Python launcher instead, for example:

python -m venv .venv

Fresh Install On macOS Or Linux

bash

mkdir privacy-filter-demo
cd privacy-filter-demo

python3 -m venv .venv
source .venv/bin/activate

python -m pip install -U pip huggingface_hub

hf download galexdav/privacy-filter-finetuned \
  --include clean_excel.py \
  --include requirements.txt \
  --include dummy_procurement_data.xlsx \
  --include label_space.json \
  --local-dir .

python -m pip install -r requirements.txt

Test The Sample Workbook

After installing, run:

python clean_excel.py dummy_procurement_data.xlsx --device cpu

This creates:

dummy_procurement_data_cleaned.xlsx

The first run downloads the checkpoint into:

models/privacy-filter-finetuned

Later runs reuse that local copy.

Use Your Own Excel File

Put your workbook in the same folder as clean_excel.py, then run:

python clean_excel.py your_data.xlsx --device cpu

Choose specific columns:

python clean_excel.py your_data.xlsx --columns Supplier Notes ContactEmail InvoiceRef Amount --device cpu

Process every sheet:

python clean_excel.py your_data.xlsx --all-sheets --device cpu

Set the output filename:

python clean_excel.py your_data.xlsx --all-sheets --device cpu --output your_data_sanitised.xlsx

Use GPU if available:

python clean_excel.py your_data.xlsx --all-sheets --device cuda

Continued Use After Installation

You do not need to reinstall each time. Open a terminal in the same folder and reactivate the environment.

Windows:

powershell

cd privacy-filter-demo
.\.venv\Scripts\Activate.ps1
python clean_excel.py another_file.xlsx --all-sheets --device cpu

macOS or Linux:

bash

cd privacy-filter-demo
source .venv/bin/activate
python clean_excel.py another_file.xlsx --all-sheets --device cpu

Important Notes For Real Data

The source workbook is not modified; the script writes a new output workbook.
The PII_Audit_Log sheet contains original detected values, so treat the cleaned workbook as sensitive if you keep that audit sheet.
Replacement consistency is currently per run, not permanent across separate days or separate executions.
If you need the same original value to always map to the same synthetic value across many files over time, add a persistent replacement map such as replacement_map.json.
This is a privacy filtering aid, not a compliance guarantee. Review outputs before production use.

Direct Python Inference

You can also use the checkpoint directly from Python:

python

from huggingface_hub import snapshot_download
from opf import OPF

checkpoint = snapshot_download("galexdav/privacy-filter-finetuned")
model = OPF(model=checkpoint, device="cpu")

result = model.redact("PO-00442 raised for Meridian Logistics Ltd, total GBP 4,250.00.")
for span in result.detected_spans:
    print(f"{span.label}: {span.text}")

Label Space

json

{
  "category_version": "custom_v1_extended",
  "span_class_names": [
    "O",
    "private_person",
    "private_email",
    "private_phone",
    "private_address",
    "account_number",
    "private_url",
    "private_date",
    "secret",
    "company_name",
    "price",
    "id_number"
  ]
}

Training Details

Field	Value
Base model	`openai/privacy-filter`
Training examples	about 830
Validation examples	about 130
Epochs	3
Hardware	1x NVIDIA L4, 24 GB, via Hugging Face Jobs
Training time	about 41 minutes

Training data was generated programmatically using templates and entity cross-products, covering all 11 categories including the original OPF categories to reduce catastrophic forgetting.

Known Limitations

Company names in isolated cells may be harder to detect without sentence context.
Price span boundaries may occasionally exclude a currency symbol.
API-key-like values can occasionally be classified as id_number rather than secret.
Performance is intended for English text.

License

Apache 2.0, same as the base model. Commercial use is permitted under the license terms.