wav2vec2-detect-toxic-th

This model is a fine-tuned version of airesearch/wav2vec2-large-xlsr-53-th on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 1.2174
• Wer: 0.4536

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: 0.0001
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 30
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.3619	3.23	100	3.2891	1.0
3.299	6.45	200	3.1670	1.0
2.1179	9.68	300	1.1747	0.5221
1.1047	12.9	400	1.0323	0.5849
0.8974	16.13	500	1.0128	0.5029
0.769	19.35	600	1.0402	0.4957
0.6659	22.58	700	1.0902	0.4729
0.6114	25.81	800	1.1412	0.4629
0.5511	29.03	900	1.1156	0.4643
0.5137	32.26	1000	1.1556	0.4679
0.5132	35.48	1100	1.1851	0.4515
0.4583	38.71	1200	1.1971	0.4529
0.4523	41.94	1300	1.2182	0.4579
0.4329	45.16	1400	1.2178	0.4586
0.4502	48.39	1500	1.2174	0.4536

Framework versions

• Transformers 4.28.0
• Pytorch 2.0.1+cu118
• Datasets 1.16.1
• Tokenizers 0.13.3