README.md

metadata

library_name: transformers
base_model: cahya/NusaBert-v1.3
tags:
  - generated_from_trainer
datasets:
  - grit-id/id_nergrit_corpus
metrics:
  - precision
  - recall
  - f1
  - accuracy
model-index:
  - name: nusabert_nergrit_1.3
    results:
      - task:
          name: Token Classification
          type: token-classification
        dataset:
          name: grit-id/id_nergrit_corpus ner
          type: grit-id/id_nergrit_corpus
          config: ner
          split: validation
          args: ner
        metrics:
          - name: Precision
            type: precision
            value: 0.8010483135824977
          - name: Recall
            type: recall
            value: 0.8338275412169375
          - name: F1
            type: f1
            value: 0.8171093159760562
          - name: Accuracy
            type: accuracy
            value: 0.9476653696498054
pipeline_tag: token-classification
license: mit
language:
  - id

NusaBert-ner-v1.3

This model is a fine-tuned version of cahya/NusaBert-v1.3 on the grit-id/id_nergrit_corpus ner dataset. It supports a context length of 8192, the same as the model cahya/NusaBert-v1.3 which was pre-trained from scratch using ModernBERT architecture. It achieves the following results on the evaluation set:

• Loss: 0.2174
• Precision: 0.8010
• Recall: 0.8338
• F1: 0.8171
• Accuracy: 0.9477

Model description

The dataset contains 19 following entities

    'CRD': Cardinal
    'DAT': Date
    'EVT': Event
    'FAC': Facility
    'GPE': Geopolitical Entity
    'LAW': Law Entity (such as Undang-Undang)
    'LOC': Location
    'MON': Money
    'NOR': Political Organization
    'ORD': Ordinal
    'ORG': Organization
    'PER': Person
    'PRC': Percent
    'PRD': Product
    'QTY': Quantity
    'REG': Religion
    'TIM': Time
    'WOA': Work of Art
    'LAN': Language

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 64
• total_eval_batch_size: 64
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 3.0

Training results

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 2.19.2
• Tokenizers 0.21.0

Usage

from transformers import pipeline
ner = pipeline("ner", model="cahya/NusaBert-ner-v1.3", grouped_entities=True)
text = "Jakarta, April 2025 - Polisi mengungkap sosok teman pemberi uang palsu kepada artis Sekar Arum Widara. Sosok tersebut ternyata adalah Bayu Setio Aribowo (BS), pegawai nonaktif Garuda yang ditangkap Polsek Tanah Abang di kasus serupa."
result = ner(text)
print(result)