PrivacyGuardML model file
==========================
Generated by Synthelion's own training pipeline:
  devtools/train_privacyguardml.py

Training data: synthetic PII values (generated from Synthelion's own
checksum validators in synthelion/privacy_validators.py and rule
patterns in synthelion/privacy_rules.yaml) inserted into real
multilingual carrier sentences (39 languages, Wikipedia corpora).
Architecture: own transformer encoder, BIO tagging head.
Runtime: fully offline, CPU-only. This is the only privacy.use_ml
backend PrivacyGuard supports — no third-party model.