Sultannn/bert-base-ft-ner-xtreme-id

This model is a fine-tuned version of bert-base-multilingual-uncased on an xtreme PAN-X.id for NER downstream task.

Details of the downstream task (NER) - Dataset

Dataset	# Examples
Train	35 K
Validation	5 K

Metrics on evaluation set

Metrics	Score
Accuracy	97.18
F1	93.26
Precision	92.36
Recall	94.18

Training hyperparameters

• Optimizer = AdamW
• LearningRate = 4e-5
• WeightDecay = 1e-2
• Warmup = 500

Example of usage

# pipeline example

from transformers import pipeline

model_checkpoint = "Sultannn/bert-base-ft-ner-xtreme-id"
token_classifier = pipeline(
    "token-classification", model=model_checkpoint, aggregation_strategy="simple")

text = "nama saya Tono saya bekerja di Facebook dan tinggal di Jawa"

token_classifier(text)

Framework versions

• Transformers 4.18.0
• TensorFlow 2.8.0
• Datasets 2.1.0
• Tokenizers 0.12.1

Fine-tune on NER script provided by @Sultan

Made with ♥ in 🌏