distilbert-base-uncased__hate_speech_offensive__train-16-1

This model is a fine-tuned version of distilbert-base-uncased on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.0424
• Accuracy: 0.5355

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: 4
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
1.0989	1.0	10	1.1049	0.1
1.0641	2.0	20	1.0768	0.3
0.9742	3.0	30	1.0430	0.4
0.8765	4.0	40	1.0058	0.4
0.6979	5.0	50	0.8488	0.7
0.563	6.0	60	0.7221	0.7
0.4135	7.0	70	0.6587	0.8
0.2509	8.0	80	0.5577	0.7
0.0943	9.0	90	0.5840	0.7
0.0541	10.0	100	0.6959	0.7
0.0362	11.0	110	0.6884	0.6
0.0254	12.0	120	0.9263	0.6
0.0184	13.0	130	0.7992	0.6
0.0172	14.0	140	0.7351	0.6
0.0131	15.0	150	0.7664	0.6
0.0117	16.0	160	0.8262	0.6
0.0101	17.0	170	0.8839	0.6
0.0089	18.0	180	0.9018	0.6

Framework versions

• Transformers 4.15.0
• Pytorch 1.10.2+cu102
• Datasets 1.18.2
• Tokenizers 0.10.3