wav2vec2-xls-r-300m-closest-to-faroese-15k-steps_v2

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: inf
• Wer: 66.0768
• Cer: 16.7567

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: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 3000
• training_steps: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.0453	0.4269	1000	inf	100.0	98.8207
0.9451	0.8538	2000	inf	65.6092	19.1500
0.6568	1.2805	3000	inf	51.9285	14.6053
0.569	1.7074	4000	inf	45.3550	11.9565
0.5004	2.1340	5000	inf	43.6086	11.7319
0.8088	2.5609	6000	inf	50.6818	13.4300
1.1499	2.9878	7000	inf	71.0149	19.1687
1.2865	3.4145	8000	inf	75.6141	19.9224
1.2107	3.8414	9000	inf	77.0186	20.9659
1.0707	4.2681	10000	inf	65.7047	17.0968
1.0988	4.6950	11000	inf	63.3710	16.1982
1.1639	5.1217	12000	inf	66.0758	16.7507
1.0514	5.5486	13000	inf	66.0777	16.7492
1.1223	5.9755	14000	inf	66.0699	16.7496
1.1473	6.4021	15000	inf	66.0768	16.7567

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0