wav2vec2-large-xls-r-300m-sw-1hr-v1

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_8_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.8004
• Wer: 0.5902
• Cer: 0.1498

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
10.7706	4.6512	100	5.0262	1.0	1.0
3.7038	9.3023	200	3.2132	1.0	1.0
2.9571	13.9535	300	2.8597	1.0	1.0
2.7859	18.6047	400	2.6007	1.0	0.7810
1.2103	23.2558	500	0.8662	0.6976	0.1859
0.3075	27.9070	600	0.7534	0.6533	0.1695
0.1911	32.5581	700	0.7585	0.6282	0.1607
0.1482	37.2093	800	0.8062	0.6340	0.1667
0.1241	41.8605	900	0.7999	0.6190	0.1605
0.1085	46.5116	1000	0.8105	0.6001	0.1524
0.0935	51.1628	1100	0.7972	0.5914	0.1502
0.0833	55.8140	1200	0.7978	0.5931	0.1505

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1