# TR Quality Pipeline: KenLM Perplexity + Slur/PII Filter + Educational-Value

> Source: https://sukruyusufkaya.com/learn/fine-tuning-cookbook/ftc-tr-quality-pipeline-kenlm-pii
> Updated: 2026-08-16T07:55:33.266Z
> Category: Fine-Tuning Cookbook (Model-by-Model)
> Module: Part IX — Türkçe-First & Yerelleştirme Mühendisliği
**TLDR:** Ham TR corpus'tan kaliteli FT data'ya: KenLM 5-gram TR perplexity (gibberish/MT artifact filter), TR slur/küfür filter, TR PII detection (TC kimlik no, telefon, e-mail), educational-value scorer (FineWeb adaptasyonu). RTX 4090'da 100GB TR corpus 4 saatte temizleme.

