clincolnoz/HateBERT-edos

final-lr2e-5-bs16-fp16-2

This model is a fine-tuned version of GroNLP/hateBERT on an https://github.com/rewire-online/edos dataset. It achieves the following results on the evaluation set:

• Loss: 0.4219
• F1 Macro: 0.8457
• F1 Weighted: 0.8868
• F1: 0.7658
• Accuracy: 0.887
• Confusion Matrix: [[2809 221] [ 231 739]]
• Confusion Matrix Norm: [[0.92706271 0.07293729] [0.23814433 0.76185567]]
• Classification Report: precision recall f1-score support 0 0.924013 0.927063 0.925535 3030.000

1 0.769792 0.761856 0.765803 970.000 accuracy 0.887000 0.887000 0.887000 0.887 macro avg 0.846902 0.844459 0.845669 4000.000 weighted avg 0.886614 0.887000 0.886800 4000.000

Model description

More information needed

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 12345
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	F1 Macro	F1 Weighted	F1	Accuracy	Confusion Matrix	Confusion Matrix Norm	Classification Report
0.3177	1.0	1000	0.2894	0.8323	0.8812	0.7373	0.886	[[2904 126]
[ 330 640]]	[[0.95841584 0.04158416]
[0.34020619 0.65979381]]	precision recall f1-score support
0 0.897959 0.958416 0.927203 3030.000
1 0.835509 0.659794 0.737327 970.000
accuracy 0.886000 0.886000 0.886000 0.886
macro avg 0.866734 0.809105 0.832265 4000.000
weighted avg 0.882815 0.886000 0.881158 4000.000
0.2232	2.0	2000	0.3370	0.8405	0.8830	0.7579	0.8832	[[2802 228]
[ 239 731]]	[[0.92475248 0.07524752]
[0.24639175 0.75360825]]	precision recall f1-score support
0 0.921407 0.924752 0.923077 3030.00000
1 0.762252 0.753608 0.757906 970.00000
accuracy 0.883250 0.883250 0.883250 0.88325
macro avg 0.841830 0.839180 0.840491 4000.00000
weighted avg 0.882812 0.883250 0.883023 4000.00000
0.1534	3.0	3000	0.4219	0.8457	0.8868	0.7658	0.887	[[2809 221]
[ 231 739]]	[[0.92706271 0.07293729]
[0.23814433 0.76185567]]	precision recall f1-score support
0 0.924013 0.927063 0.925535 3030.000
1 0.769792 0.761856 0.765803 970.000
accuracy 0.887000 0.887000 0.887000 0.887
macro avg 0.846902 0.844459 0.845669 4000.000
weighted avg 0.886614 0.887000 0.886800 4000.000

Framework versions

• Transformers 4.27.0.dev0
• Pytorch 1.13.1+cu117
• Datasets 2.9.0
• Tokenizers 0.13.2