Downloads · 30 days
32
14% of all-time downloads
paragdakle/greek-ner-text-wikiann-greekbert
greek-ner-text-wikiann-greekbert is a token classification model from paragdakle. Use it when you need labels on individual words, such as names. The card lists the license as apache-2.0.
Textual-entity member of the FinNLP 2026 Subtask 1 (GRFinNER) submission's ensemble -- supplies ΠΡΟΣΩΠΟ only, see training data note. Fine-tuned nlpaueb/bert-base-greek-uncased-v1 on WikiANN-Greek.
Downloads · 30 days
32
14% of all-time downloads
All-time downloads
235
Public
Parameters
112M
449 MB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors449 MB · 100%
From the Hugging Face model README
Textual-entity member of the FinNLP 2026 Subtask 1 (GRFinNER) submission's ensemble -- supplies ΠΡΟΣΩΠΟ only, see training data note. Fine-tuned nlpaueb/bert-base-greek-uncased-v1 on WikiANN-Greek.
Labels: ΠΡΟΣΩΠΟ (person) / ΟΡΓΑΝΙΣΜΟΣ (organization) / ΤΟΠΟΘΕΣΙΑ (location)
import torch
from transformers import AutoModelForTokenClassification, AutoTokenizer
model_id = "paragdakle/greek-ner-text-wikiann-greekbert"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForTokenClassification.from_pretrained(model_id).eval()
text = "Στις 08.11.2019, η ΟΠΑΠ INVESTMENT LTD ήρθε σε συμφωνία..."
enc = tokenizer(text, return_offsets_mapping=True, return_tensors="pt")
offsets = enc.pop("offset_mapping")[0]
with torch.no_grad():
logits = model(**enc).logits
tags = [model.config.id2label[i] for i in logits.argmax(-1)[0].tolist()]
# pair `tags` with `offsets` to recover character-span entities (see the
# submission's ner_utils.encoder_predict for the full decode)
20,000 sentences from unimelb-nlp/wikiann (config 'el') -- general-domain Greek Wikipedia text, silver-standard (auto-generated from hyperlinks, not hand-annotated). Used because no labeled Greek person/org/location data for the financial-report domain is published anywhere we could find. In the submitted ensemble this model supplies ΠΡΟΣΩΠΟ only -- an ablation showed its ΟΡΓΑΝΙΣΜΟΣ/ΤΟΠΟΘΕΣΙΑ predictions transfer poorly to financial-report text (dropped legal suffixes, missed foreign company names) despite the strong WikiANN-domain score above.