bert-finetuned-ner-ontonotes

This model is a fine-tuned version of bert-base-cased on the ontonotes5 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1503
• Precision: 0.8567
• Recall: 0.8842
• F1: 0.8702
• Accuracy: 0.9755

Model description

Token classification experiment, NER, on business topics.

Intended uses & limitations

The model can be used on token classification, in particular NER. It is fine tuned on business topic.

Training and evaluation data

The dataset used is ontonotes5

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 6

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.0842	1.0	7491	0.0950	0.8524	0.8715	0.8618	0.9745
0.0523	2.0	14982	0.1044	0.8449	0.8827	0.8634	0.9744
0.036	3.0	22473	0.1118	0.8529	0.8843	0.8683	0.9760
0.0231	4.0	29964	0.1240	0.8589	0.8805	0.8696	0.9752
0.0118	5.0	37455	0.1416	0.8570	0.8804	0.8685	0.9753
0.0077	6.0	44946	0.1503	0.8567	0.8842	0.8702	0.9755

Framework versions

• Transformers 4.22.1
• Pytorch 1.12.1+cu113
• Datasets 2.5.1
• Tokenizers 0.12.1