wav2vec2-base-vios-commonvoice

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3823
• Wer: 0.2401

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.2268	0.66	500	0.8746	0.5939
0.8728	1.32	1000	0.6435	0.4554
0.6899	1.99	1500	0.5655	0.3995
0.5842	2.65	2000	0.5267	0.3694
0.5371	3.31	2500	0.4980	0.3431
0.4921	3.97	3000	0.4781	0.3276
0.4508	4.64	3500	0.4434	0.3134
0.433	5.3	4000	0.4348	0.2963
0.404	5.96	4500	0.4248	0.2874
0.3834	6.62	5000	0.4163	0.2775
0.3784	7.28	5500	0.4104	0.2751
0.3669	7.95	6000	0.4143	0.2724
0.3462	8.61	6500	0.4131	0.2699
0.3364	9.27	7000	0.4070	0.2617
0.3249	9.93	7500	0.4076	0.2603
0.3154	10.6	8000	0.3998	0.2577
0.3117	11.26	8500	0.3930	0.2505
0.3101	11.92	9000	0.4003	0.2492
0.298	12.58	9500	0.3960	0.2496
0.2968	13.24	10000	0.3877	0.2469
0.29	13.91	10500	0.3870	0.2456
0.2921	14.57	11000	0.3823	0.2401

Framework versions

• Transformers 4.19.2
• Pytorch 1.11.0+cu113
• Datasets 2.2.2
• Tokenizers 0.12.1