metadata

tags:
  - generated_from_trainer
datasets:
  - null
metrics:
  - precision
  - recall
  - f1
  - accuracy
language:
  - sr
model_index:
  - name: bert-srb-ner
    results:
      - task:
          name: Token Classification
          type: token-classification
        metric:
          name: Accuracy
          type: accuracy
          value: 0.9641060273510046

bert-srb-ner

This model was finetuned from Aleksandar/bert-srb-cased-oscar on the setimes.SR dataset. It achieves the following results on the evaluation set:

Loss: 0.1647
Precision: 0.8247
Recall: 0.8454
F1: 0.8349
Accuracy: 0.9641

Model description

Default settings for BERT model, finetuned with batch size of 16.

Intended uses & limitations

Tag (IOB)	Numerical representation	Meaning (Beginning = B., Inside = I.)
O	0	Other
B-per	1	B.Person
I-per	2	I. Person
B-org	3	B. organization
I-org	4	I. organization
B-loc	5	B. location
I-loc	6	I. location
B-misc	7	B. Miscellaneous
I-misc	8	I. Miscellaneous
B-deriv-per	9	B. Derived Person

MIT license

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 16
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	1.0	207	0.2040	0.7006	0.7466	0.7228	0.9411
No log	2.0	414	0.1561	0.7299	0.7868	0.7573	0.9519
0.2313	3.0	621	0.1455	0.7693	0.7992	0.7840	0.9567
0.2313	4.0	828	0.1628	0.7760	0.8037	0.7896	0.9570
0.0828	5.0	1035	0.1424	0.7997	0.8299	0.8145	0.9604
0.0828	6.0	1242	0.1512	0.7983	0.8361	0.8168	0.9618
0.0828	7.0	1449	0.1587	0.8084	0.8415	0.8246	0.9627
0.0362	8.0	1656	0.1613	0.8154	0.8358	0.8255	0.9632
0.0362	9.0	1863	0.1685	0.8211	0.8429	0.8319	0.9632
0.0174	10.0	2070	0.1647	0.8247	0.8454	0.8349	0.9641

Framework versions

Transformers 4.9.2
Pytorch 1.9.0
Datasets 1.11.0
Tokenizers 0.10.1