QomSSLab/etghan_tagging_v2

This repository hosts an XLM-RoBERTa token-classification head trained.

Usage

from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

model_id = "QomSSLab/etghan_tagging_v2"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForTokenClassification.from_pretrained(model_id)
tagger = pipeline("token-classification", model=model, tokenizer=tokenizer, aggregation_strategy="simple")

text = "مثال از یک ورودی فارسی"
for entity in tagger(text):
    print(entity)

Labels

  • ABSENCE
  • APPEALABILITY
  • APPEAL_AUTHORITY
  • APPEAL_DEADLINE
  • COURT_NAME
  • DEFENDANT_ADDRESS
  • DEFENDANT_FATHER
  • DEFENDANT_NAME_LEGAL
  • DEFENDANT_NAME_NATURAL
  • FINALITY
  • JUDGE_NAME
  • JUDGE_POSITION
  • JUDGE_SIGNATURE
  • JUDGMENT_DATE
  • LEGAL_CHARGE
  • LEGAL_REPRESENTATIVE
  • O
  • PLACE_OF_OFFENSE
  • PLAINTIFF_ADDRESS
  • PLAINTIFF_CLAIM
  • PLAINTIFF_FATHER
  • PLAINTIFF_NAME_LEGAL
  • PLAINTIFF_NAME_NATURAL
  • PRESENCE
  • SUBJECT_OF_CLAIM
  • TIME_OF_OFFENSE

Metrics

Validation Metrics

  • Precision: 0.8207
  • Recall: 0.7751
  • F1: 0.7972
  • Accuracy: 0.9701

Per-label Breakdown

Label Precision Recall F1 Support
ABSENCE 0.8947 0.6071 0.7234 56
APPEALABILITY 0.9423 0.9976 0.9691 409
APPEAL_AUTHORITY 0.9748 0.9884 0.9816 431
APPEAL_DEADLINE 0.8976 0.9544 0.9252 395
COURT_NAME 0.5248 0.9815 0.6839 54
DEFENDANT_ADDRESS 0.9586 0.9929 0.9754 140
DEFENDANT_FATHER 0.9672 0.9866 0.9768 299
DEFENDANT_NAME_LEGAL 0.9101 1.0000 0.9529 81
DEFENDANT_NAME_NATURAL 0.9767 0.9254 0.9504 724
FINALITY 0.8056 0.8529 0.8286 34
JUDGE_NAME 0.9534 0.9840 0.9685 312
JUDGE_POSITION 1.0000 0.9615 0.9804 260
JUDGE_SIGNATURE 1.0000 1.0000 1.0000 0
JUDGMENT_DATE 0.9409 0.9835 0.9618 243
LEGAL_CHARGE 1.0000 0.9343 0.9660 700
LEGAL_REPRESENTATIVE 0.9552 0.8114 0.8774 525
O 0.9774 0.9912 0.9843 68829
PLACE_OF_OFFENSE 0.6667 0.4000 0.5000 10
PLAINTIFF_ADDRESS 0.9375 0.9740 0.9554 77
PLAINTIFF_CLAIM 0.8740 0.6617 0.7532 2779
PLAINTIFF_FATHER 0.9757 0.9757 0.9757 247
PLAINTIFF_NAME_LEGAL 0.7250 1.0000 0.8406 29
PLAINTIFF_NAME_NATURAL 0.9679 0.9679 0.9679 560
PRESENCE 0.9812 0.9573 0.9691 164
SUBJECT_OF_CLAIM 0.8039 0.6726 0.7324 1457
TIME_OF_OFFENSE 0.8148 0.4681 0.5946 47
Downloads last month
32
Safetensors
Model size
0.6B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support