wav2vec2-xls-r-1b-distant-from-faroese-251h-30-epochs_20250118_v3

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 4.5739
• Wer: 100.8689
• Cer: 65.0005

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 OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 3000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.6412	0.6723	1000	3.3884	100.0	100.0
2.713	1.3442	2000	1.5597	99.8034	59.2021
1.7455	2.0161	3000	1.1254	97.2259	36.0451
1.5575	2.6884	4000	1.0471	96.9468	30.6516
1.2979	3.3603	5000	1.0094	96.4298	30.0615
1.6143	4.0323	6000	1.1377	100.0437	35.8721
2.0126	4.7045	7000	1.4830	99.8204	54.5017
2.3025	5.3765	8000	1.8700	99.9903	72.4594
2.3062	6.0484	9000	1.9594	99.7452	53.2416
2.3507	6.7207	10000	1.9187	99.9757	63.1540
2.3521	7.3926	11000	1.8744	99.9927	71.1402
2.32	8.0645	12000	1.8584	99.9976	69.5654
2.64	8.7368	13000	1.8953	100.0	60.8263
3.3038	9.4087	14000	2.5665	99.8738	54.2732
3.9447	10.0807	15000	3.4153	109.2665	64.7511
5.0653	10.7529	16000	4.1121	110.7956	64.6105
4.7627	11.4249	17000	4.3688	130.6878	62.1256
5.154	12.0968	18000	4.5150	101.0315	64.4541
5.617	12.7691	19000	4.5738	100.8786	65.0001
5.1644	13.4410	20000	4.5739	100.8592	64.9922
6.064	14.1129	21000	4.5739	100.8568	65.0091
5.4595	14.7852	22000	4.5739	100.8737	64.9893
5.6081	15.4571	23000	4.5739	100.8762	64.9944
5.4959	16.1291	24000	4.5738	100.8713	65.0048
5.8105	16.8013	25000	4.5738	100.8640	64.9987
5.8077	17.4733	26000	4.5737	100.8640	65.0063
5.6071	18.1452	27000	4.5740	100.8786	65.0142
5.0926	18.8175	28000	4.5738	100.8689	64.9947
5.6166	19.4894	29000	4.5739	100.8689	64.9929
5.3716	20.1613	30000	4.5740	100.8665	65.0034
5.6498	20.8336	31000	4.5738	100.8737	65.0138
5.27	21.5055	32000	4.5740	100.8665	64.9998
5.3673	22.1775	33000	4.5738	100.8786	65.0106
5.6269	22.8497	34000	4.5738	100.8762	65.0037
5.8331	23.5217	35000	4.5738	100.8737	65.0066
5.8557	24.1936	36000	4.5739	100.8689	65.0088
5.4928	24.8659	37000	4.5739	100.8689	65.0070
4.9825	25.5378	38000	4.5738	100.8713	64.9994
5.5315	26.2097	39000	4.5738	100.8616	64.9972
5.7108	26.8820	40000	4.5738	100.8737	65.0048
5.977	27.5539	41000	4.5737	100.8835	64.9990
5.9164	28.2259	42000	4.5740	100.8737	64.9983
5.7156	28.8982	43000	4.5739	100.8713	65.0041
5.2768	29.5701	44000	4.5739	100.8689	65.0005

Framework versions

• Transformers 4.48.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0