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: 5000
num_epochs: 30
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.3305	0.4877	1000	3.2645	100.0	97.5974
1.1704	0.9754	2000	0.7925	67.4671	19.9235
0.5938	1.4628	3000	0.3347	40.9349	10.8308
0.5005	1.9505	4000	0.2706	36.3352	9.2638
0.3793	2.4379	5000	0.2100	31.7795	7.8988
0.3445	2.9256	6000	0.1875	30.1141	7.4262
0.2687	3.4131	7000	0.1650	27.4926	6.7035
0.2772	3.9008	8000	0.1514	26.4661	6.3634
0.222	4.3882	9000	0.1522	26.3251	6.2514
0.2334	4.8759	10000	0.1476	25.3073	6.0399
0.1898	5.3633	11000	0.1410	25.0165	5.8947
0.1943	5.8510	12000	0.1306	24.8755	5.7945
0.1694	6.3385	13000	0.1276	24.1089	5.6967
0.1768	6.8261	14000	0.1249	23.5890	5.4955
0.1471	7.3136	15000	0.1254	23.7829	5.5286
0.1542	7.8013	16000	0.1181	22.9590	5.3069
0.1403	8.2887	17000	0.1245	22.8797	5.2462
0.1457	8.7764	18000	0.1170	22.8268	5.1925
0.1195	9.2638	19000	0.1170	22.5052	5.1665
0.1231	9.7515	20000	0.1180	22.4127	5.0994
0.1109	10.2390	21000	0.1197	22.0734	5.0103
0.1141	10.7267	22000	0.1138	21.9104	4.9613
0.0922	11.2141	23000	0.1190	21.7033	4.9424
0.0935	11.7018	24000	0.1155	21.6857	4.9108
0.0959	12.1892	25000	0.1105	21.5139	4.8422
0.0874	12.6769	26000	0.1138	21.5403	4.8643
0.0877	13.1644	27000	0.1090	21.1570	4.7412
0.082	13.6520	28000	0.1137	21.3552	4.8383
0.078	14.1395	29000	0.1144	21.1261	4.7562
0.0899	14.6272	30000	0.1077	20.8882	4.6442
0.0758	15.1146	31000	0.1125	20.8398	4.6473
0.074	15.6023	32000	0.1087	20.8882	4.6378
0.0763	16.0897	33000	0.1062	20.6018	4.5724
0.0653	16.5774	34000	0.1145	20.6679	4.5952
0.0646	17.0649	35000	0.1036	20.2097	4.4706
0.0555	17.5525	36000	0.1034	20.0775	4.4130
0.0628	18.0400	37000	0.1024	20.0687	4.3846
0.0567	18.5277	38000	0.1039	20.1304	4.4240
0.0568	19.0151	39000	0.1037	20.0115	4.3893
0.0535	19.5028	40000	0.1063	19.9938	4.4098
0.0502	19.9905	41000	0.1047	19.8969	4.3901
0.0518	20.4779	42000	0.1059	19.8176	4.3506
0.0426	20.9656	43000	0.1029	19.6105	4.2875
0.04	21.4531	44000	0.1057	19.6898	4.3112
0.0634	21.9407	45000	0.1037	19.6942	4.2907
0.0506	22.4282	46000	0.1025	19.6193	4.2725
0.0583	22.9159	47000	0.1031	19.6942	4.2725
0.0539	23.4033	48000	0.1053	19.5753	4.2457
0.0439	23.8910	49000	0.1044	19.5444	4.2638
0.0473	24.3784	50000	0.1036	19.4607	4.2165
0.0467	24.8661	51000	0.1051	19.4916	4.2347
0.0438	25.3536	52000	0.1008	19.3197	4.1794
0.0378	25.8413	53000	0.1024	19.3770	4.1913
0.0435	26.3287	54000	0.1034	19.4079	4.1913
0.0412	26.8164	55000	0.1016	19.3506	4.1920
0.0432	27.3038	56000	0.1024	19.3902	4.1991
0.0501	27.7915	57000	0.1026	19.3946	4.1952
0.0496	28.2790	58000	0.1037	19.4079	4.2062
0.0377	28.7666	59000	0.1031	19.3902	4.2055
0.0514	29.2541	60000	0.1029	19.3946	4.2023
0.047	29.7418	61000	0.1028	19.3990	4.2031

Framework versions

Transformers 4.48.1
Pytorch 2.5.1+cu121
Datasets 3.2.0
Tokenizers 0.21.0

davidilag/wav2vec2-xls-r-300m-faroese-100h-30-epochs_20250126_v2

wav2vec2-xls-r-300m-faroese-100h-30-epochs_20250126_v2

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

davidilag/
wav2vec2-xls-r-300m-faroese-100h-30-epochs_20250126_v2