This model is a proof of concept where regular LLM are finetuned to anonimise data. Using a systemprompt:
Anonymize :
and a user prompt the LLM will try to replace sensitive data by the character "^"
Targets for the anonimisation
The model should be able to recognise:
- Addresses
- emails
- phone numbers
- Religious events
- Gendre
- IBAN numbers
- Social security numbers
- Phone numbers
- Bban numbers
- swift numbers
- a small size of company
- credit cards
- passports
- name
Example
Input
My name is Mike Smith I live in Texas in the Street Rosenberg, Texas(TX), 77471. You can contact me over at mike@smith.com or via phone +13548783461.
Expected output
My name is ^^^^ ^^^^^ I live in Texas in the Street ^^^^^^^^^, ^^^^^^^^^, ^^^^^. You can contact me over at ^^^^@^^^^^.com or via phone +^^^^^^^^^^^.