distilbert-base-uncased-finetuned-ner

This model is a fine-tuned version of distilbert-base-uncased on the privy dataset. It achieves the following results on the evaluation set:

Loss: 0.0016
Precision: 0.9984
Recall: 0.9986
F1: 0.9985
Accuracy: 0.9996

Model description

Output indices map to following labels:

['O', 'B-O', 'I-O', 'L-O', 'U-O', 'B-PER', 'I-PER', 'L-PER', 'U-PER', 'B-LOC', 'I-LOC', 'L-LOC', 'U-LOC', 'B-ORG', 'I-ORG', 'L-ORG', 'U-ORG', 'B-NRP', 'I-NRP', 'L-NRP', 'U-NRP', 'B-DATE_TIME', 'I-DATE_TIME', 'L-DATE_TIME', 'U-DATE_TIME', 'B-CREDIT_CARD', 'I-CREDIT_CARD', 'L-CREDIT_CARD', 'U-CREDIT_CARD', 'B-URL', 'I-URL', 'L-URL', 'U-URL', 'B-IBAN_CODE', 'I-IBAN_CODE', 'L-IBAN_CODE', 'U-IBAN_CODE', 'B-US_BANK_NUMBER', 'I-US_BANK_NUMBER', 'L-US_BANK_NUMBER', 'U-US_BANK_NUMBER', 'B-PHONE_NUMBER', 'I-PHONE_NUMBER', 'L-PHONE_NUMBER', 'U-PHONE_NUMBER', 'B-US_SSN', 'I-US_SSN', 'L-US_SSN', 'U-US_SSN', 'B-US_PASSPORT', 'I-US_PASSPORT', 'L-US_PASSPORT', 'U-US_PASSPORT', 'B-US_DRIVER_LICENSE', 'I-US_DRIVER_LICENSE', 'L-US_DRIVER_LICENSE', 'U-US_DRIVER_LICENSE', 'B-US_LICENSE_PLATE', 'I-US_LICENSE_PLATE', 'L-US_LICENSE_PLATE', 'U-US_LICENSE_PLATE', 'B-IP_ADDRESS', 'I-IP_ADDRESS', 'L-IP_ADDRESS', 'U-IP_ADDRESS', 'B-US_ITIN', 'I-US_ITIN', 'L-US_ITIN', 'U-US_ITIN', 'B-EMAIL_ADDRESS', 'I-EMAIL_ADDRESS', 'L-EMAIL_ADDRESS', 'U-EMAIL_ADDRESS', 'B-TITLE', 'I-TITLE', 'L-TITLE', 'U-TITLE', 'B-COORDINATE', 'I-COORDINATE', 'L-COORDINATE', 'U-COORDINATE', 'B-IMEI', 'I-IMEI', 'L-IMEI', 'U-IMEI', 'B-PASSWORD', 'I-PASSWORD', 'L-PASSWORD', 'U-PASSWORD', 'B-LICENSE_PLATE', 'I-LICENSE_PLATE', 'L-LICENSE_PLATE', 'U-LICENSE_PLATE', 'B-CURRENCY', 'I-CURRENCY', 'L-CURRENCY', 'U-CURRENCY', 'B-FINANCIAL', 'I-FINANCIAL', 'L-FINANCIAL', 'U-FINANCIAL', 'B-ROUTING_NUMBER', 'I-ROUTING_NUMBER', 'L-ROUTING_NUMBER', 'U-ROUTING_NUMBER', 'B-SWIFT_CODE', 'I-SWIFT_CODE', 'L-SWIFT_CODE', 'U-SWIFT_CODE', 'B-MAC_ADDRESS', 'I-MAC_ADDRESS', 'L-MAC_ADDRESS', 'U-MAC_ADDRESS', 'B-AGE', 'I-AGE', 'L-AGE', 'U-AGE']

Intended uses & limitations

NER detection for PII anonymization

Training and evaluation data

beki/privy dataset

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 16
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.0028	1.0	6310	0.0025	0.9977	0.9977	0.9977	0.9995
0.0015	2.0	12620	0.0017	0.9983	0.9985	0.9984	0.9996
0.001	3.0	18930	0.0016	0.9984	0.9986	0.9985	0.9996

Framework versions

Transformers 4.30.2
Pytorch 2.0.1+cu118
Datasets 2.13.1
Tokenizers 0.13.3