turkish-pii-detection-GGUF

What is this model? turkish-pii-detection-GGUF is the GGUF (BF16, Q8_0) build of halilneed/turkish-pii-detection for local PII masking of Turkish text with Ollama or llama.cpp on CPU, without Python or transformers.

EN: Turkish PII detection / KVKK data masking, offline โ€” GGUF (BF16, Q8_0) build for Ollama and llama.cpp. TR: Tรผrkรงe kiลŸisel veri tespiti / KVKK veri maskeleme, รงevrimdฤฑลŸฤฑ โ€” Ollama ve llama.cpp iรงin GGUF (BF16, Q8_0) sรผrรผmรผ. Model page & setup help / Model sayfasฤฑ ve kurulum desteฤŸi: halilneed.agency/models/turkish-pii-detection

halilneed/turkish-pii-detection v02'nin GGUF sรผrรผmรผ: Ollama ve llama.cpp ile yerelde, Python ve transformers olmadan รงalฤฑลŸฤฑr. Tรผrkรงe metindeki kiลŸisel verileri (ad, TCKN, IBAN, telefon, e-posta, adres ve 53 etiketin geri kalanฤฑ) metin bir LLM'e ya da log deposuna gitmeden รถnce etiketle deฤŸiลŸtirir; maskeleme politikasฤฑnฤฑ talimat olarak okur.

Files / Dosyalar: turkish-pii-detection-Q8_0.gguf 299.7 MB (299,748,544 bytes) ยท turkish-pii-detection-BF16.gguf 551.0 MB (551,038,144 bytes). Runs on CPU, no GPU needed (llama.cpp / Ollama) / GPU gerekmez, CPU'da รงalฤฑลŸฤฑr. Try the browser demo / Tarayฤฑcฤฑ demosunu deneyin (demo runs the base model's ONNX conversion, not these GGUF files / demo bu GGUF dosyalarฤฑnฤฑ deฤŸil ana modelin ONNX dรถnรผลŸรผmรผnรผ รงalฤฑลŸtฤฑrฤฑr).

Dosya Boyut Benchmark (1.000 satฤฑr, tam eลŸleลŸme) Ana modelle aynฤฑ รงฤฑktฤฑ
ana model, transformers BF16 536 MB 0.944 โ€”
turkish-pii-detection-BF16.gguf 551 MB 0.943 %99,6
turkish-pii-detection-Q8_0.gguf 300 MB 0.944 %98,9

ร–lรงรผm: cagrigungor/turkish-pii-masking-benchmark, satฤฑr dรผzeyi tam eลŸleลŸme (strip sonrasฤฑ karakter karakter aynฤฑ), ana model kartฤฑndaki betikle aynฤฑ metrik. GGUF'lar Ollama 0.40.7 รผzerinden, sฤฑcaklฤฑk 0 ile koลŸuldu. Model tahminini yaptฤฑฤŸฤฑ 1.000 satฤฑrฤฑn 996'sฤฑnda (BF16) ve 989'unda (Q8_0) ana modelle birebir aynฤฑ รงฤฑktฤฑyฤฑ verdi.

Q8_0'ฤฑn zayฤฑf dilimleri ana modelle aynฤฑ: ek almฤฑลŸ PII 0.803, BรœYรœK HARF 0.847. Tek fark, 20 satฤฑrlฤฑk รงok kiลŸili metin diliminde: 0.650, ana modelde 0.750 (iki satฤฑr). ร‡ok kiลŸili metinleri maskeliyorsan BF16'yฤฑ kullan.

Q4_K_M yok: modelin 268M parametresinin 168M'i embedding tablosunda ve Q8_0 zaten 300 MB. 4 bitlik sรผrรผmรผ kรผรงรผk bir modelde kaybฤฑnฤฑ รถlรงmeden yayฤฑnlamฤฑyorum.

Ollama

ollama pull hf.co/halilneed/turkish-pii-detection-GGUF:Q8_0

Deterministik รงฤฑktฤฑ ve doฤŸru durma iรงin bu repodaki Modelfile ile kendi adฤฑnฤฑ ver:

# Modelfile ve .gguf aynฤฑ klasรถrde
ollama create turkish-pii -f Modelfile
ollama run turkish-pii "Metindeki tรผm kiลŸisel verileri uygun etiketlerle maskele.

Metin: mรผลŸteri AyลŸe Yฤฑlmaz tc 12345678950 tel 0532 111 22 33"
# -> mรผลŸteri [AD] tc [TCKN] tel [TEL]

ฤฐstek her zaman <talimat>, boลŸ satฤฑr, Metin: <metin> biรงimindedir. Talimat politikayฤฑ belirler: "Sadece telefon numaralarฤฑnฤฑ maskele", "TCKN dฤฑลŸฤฑndaki her ลŸeyi maskele" gibi. Politikalar ve 53 etiketlik ลŸema ana model kartฤฑnda.

Python'dan, Ollama'nฤฑn yerel API'siyle:

import json, urllib.request

def maskele(metin, talimat="Metindeki tรผm kiลŸisel verileri uygun etiketlerle maskele."):
    body = {"model": "turkish-pii", "prompt": f"{talimat}\n\nMetin: {metin}", "stream": False,
            "options": {"temperature": 0}}
    req = urllib.request.Request("http://127.0.0.1:11434/api/generate", json.dumps(body).encode(),
                                 {"Content-Type": "application/json"})
    return json.loads(urllib.request.urlopen(req).read())["response"].strip()

llama.cpp

llama-cli -hf halilneed/turkish-pii-detection-GGUF:Q8_0 --temp 0 -n 512

DรถnรผลŸtรผrme

  • llama.cpp convert_hf_to_gguf.py (commit 1ab09d6), --outtype bf16 ve --outtype q8_0, ana modelin v02 aฤŸฤฑrlฤฑฤŸฤฑndan (sha256 Hub'daki model.safetensors ile aynฤฑ).
  • Ana modelin tokenizer'ฤฑnda embedding tablosunda karลŸฤฑlฤฑฤŸฤฑ olmayan bir token var: Gemma 3'รผn gรถrsel tokenฤฑ <image_soft_token> (id 262144). DรถnรผลŸtรผrรผcรผ bunu kabul etmediฤŸi iรงin dรถnรผลŸtรผrme kopyasฤฑndan รงฤฑkarฤฑldฤฑ. Benchmark'ฤฑn 1.000 promptu ve 1.000 beklenen รงฤฑktฤฑsฤฑnda tokenizasyon birebir aynฤฑ kaldฤฑ.
  • Ollama 0.40.7 safetensors iรงe aktarmada Gemma 3'รผ desteklemiyor; bu yรผzden GGUF.

Sฤฑnฤฑrlar

Ana modelin sฤฑnฤฑrlarฤฑ geรงerli: sentetik veriyle deฤŸerlendirildi, รผretimden รถnce kendi metninde รถlรง; baฤŸlamsฤฑz kฤฑsa parรงalarda temkinli; ลŸema dฤฑลŸฤฑ etiket รผretebilir, รงฤฑktฤฑyฤฑ etiket listesine karลŸฤฑ doฤŸrula. KVKK/GDPR uyumu kullananฤฑn sorumluluฤŸundadฤฑr. Lisans: Gemma Terms of Use.

English

GGUF build of halilneed/turkish-pii-detection v02, a 270M model that masks Turkish personal data (names, national ID numbers, IBANs, phone numbers, emails, addresses; 53 labels) following a Turkish instruction. Runs locally with Ollama or llama.cpp. On the 1,000-row cagrigungor/turkish-pii-masking-benchmark, Q8_0 (300 MB) scores 0.944 row-level exact match, the same as the original model, and BF16 scores 0.943. The one difference is the 20-row multi-person slice, where Q8_0 scores 0.650 against 0.750. The prompt is <instruction>, a blank line, then Metin: <text>; use the Modelfile here for temperature 0 and the right stop token.

Model page with every result: halilneed.agency/models/turkish-pii-detection

Related models / ฤฐlgili modeller

All five in one collection / Hepsi tek koleksiyonda: Turkish Privacy & Safety Models

FAQ / Sฤฑk sorulan sorular

Which file should I use? / Hangi dosyayฤฑ kullanmalฤฑyฤฑm? Q8_0 (300 MB) scores 0.944 row-level exact match on the 1,000-row benchmark, same as the original model; BF16 (551 MB) scores 0.943. On the 20-row multi-person slice Q8_0 scores 0.650 vs 0.750, so use BF16 for multi-person texts. / ร‡ok kiลŸili metinler iรงin BF16 รถnerilir.

Is there a Q4_K_M build? / Q4_K_M sรผrรผmรผ var mฤฑ? No. 168M of the 268M parameters are in the embedding table and Q8_0 is already 300 MB; a 4-bit build is not published without measuring its loss. / Hayฤฑr, kaybฤฑ รถlรงรผlmeden yayฤฑnlanmฤฑyor.

What is the prompt format? / ฤฐstek biรงimi nedir? <instruction>, a blank line, then Metin: <text>. Use the Modelfile in this repo for temperature 0 and the right stop token. / Talimat maskeleme politikasฤฑnฤฑ belirler.

Does it guarantee KVKK/GDPR compliance? / KVKK/GDPR uyumu saฤŸlar mฤฑ? No. It was evaluated on synthetic data; measure on your own text, validate labels against the schema. Compliance is the user's responsibility. / Uyum kullananฤฑn sorumluluฤŸundadฤฑr.

Resources / Kaynaklar

Downloads last month
1,137
GGUF
Model size
0.3B params
Architecture
gemma3
Hardware compatibility
Log In to add your hardware

8-bit

16-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for halilneed/turkish-pii-detection-GGUF

Collection including halilneed/turkish-pii-detection-GGUF