distilbert-base-uncased__hate_speech_offensive__train-16-6

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: 0.8331
• Accuracy: 0.625

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
1.0881	1.0	10	1.1248	0.1
1.0586	2.0	20	1.1162	0.2
0.9834	3.0	30	1.1199	0.3
0.9271	4.0	40	1.0740	0.3
0.7663	5.0	50	1.0183	0.5
0.6042	6.0	60	1.0259	0.5
0.4482	7.0	70	0.8699	0.7
0.3072	8.0	80	1.0615	0.5
0.1458	9.0	90	1.0164	0.5
0.0838	10.0	100	1.0620	0.5
0.055	11.0	110	1.1829	0.5
0.0347	12.0	120	1.2815	0.4
0.0244	13.0	130	1.2607	0.6
0.0213	14.0	140	1.3695	0.5
0.0169	15.0	150	1.4397	0.5
0.0141	16.0	160	1.4388	0.6
0.0122	17.0	170	1.4242	0.6

Framework versions

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