wav2vec2-xls-r-300m-closest-to-faroese-full-15k-steps_v7

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: 1.2472
• Wer: 95.4633
• Cer: 32.3403

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: 0.0003
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 16
• 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: 500
• training_steps: 15000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
13.1629	0.3349	1000	0.6152	63.3285	18.8424
10.2766	0.6698	2000	0.4313	49.4623	13.6814
7.9374	1.0044	3000	0.3649	43.9210	11.8791
7.6787	1.3392	4000	0.3353	40.3217	10.5091
7.4414	1.6741	5000	0.3097	38.5529	10.0199
5.6641	2.0087	6000	0.2806	37.1760	9.5132
19.8102	2.3436	7000	1.0119	81.0784	25.2870
21.6448	2.6785	8000	1.1365	77.7051	26.4944
20.3363	3.0131	9000	1.0859	88.1525	32.7446
20.0312	3.3479	10000	1.0513	88.4973	30.9106
21.5903	3.6828	11000	1.0996	90.2825	32.8604
21.0909	4.0174	12000	1.0407	84.0173	26.3710
23.4511	4.3523	13000	1.2414	80.7754	24.7467
23.5201	4.6872	14000	1.2473	95.4625	32.3366
24.6061	5.0218	15000	1.2472	95.4633	32.3403

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0