torgo_xlsr_finetune_M02

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

• Loss: 1.4526
• Wer: 0.2394

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.5091	0.85	1000	3.2824	1.0
1.6986	1.7	2000	1.6782	0.7538
0.8367	2.55	3000	1.1374	0.5772
0.5881	3.4	4000	1.2548	0.4567
0.5024	4.26	5000	1.1946	0.3905
0.4208	5.11	6000	1.3830	0.3939
0.3647	5.96	7000	1.2722	0.3404
0.3357	6.81	8000	1.1639	0.3506
0.2972	7.66	9000	1.3775	0.3192
0.27	8.51	10000	1.2345	0.3200
0.2532	9.36	11000	1.1760	0.3158
0.2171	10.21	12000	1.3310	0.2980
0.2367	11.06	13000	1.2634	0.2912
0.2086	11.91	14000	1.3451	0.2742
0.2023	12.77	15000	1.4799	0.2912
0.1862	13.62	16000	1.7212	0.2869
0.1698	14.47	17000	1.4304	0.2725
0.1489	15.32	18000	1.6233	0.2759
0.1489	16.17	19000	1.6320	0.2657
0.1552	17.02	20000	1.3667	0.2377
0.1414	17.87	21000	1.5199	0.2470
0.1195	18.72	22000	1.4951	0.2368
0.1303	19.57	23000	1.4526	0.2394

Framework versions

• Transformers 4.26.1
• Pytorch 2.2.1
• Datasets 2.18.0
• Tokenizers 0.13.3