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_FUSED 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
3.3527	0.4877	1000	3.2659	100.0	98.2563
0.874	0.9754	2000	0.5326	46.3057	12.6393
0.3969	1.4628	3000	0.2294	29.6956	7.6069
0.3571	1.9505	4000	0.1840	27.9949	6.8731
0.2867	2.4379	5000	0.1718	26.4440	6.4778
0.2678	2.9256	6000	0.1556	25.2148	6.1015
0.1909	3.4131	7000	0.1428	24.4261	5.8206
0.2102	3.9008	8000	0.1364	24.0208	5.6146
0.1689	4.3882	9000	0.1316	23.2982	5.4632
0.1738	4.8759	10000	0.1323	22.7739	5.3661
0.1432	5.3633	11000	0.1426	23.2850	5.5089
0.1489	5.8510	12000	0.1308	22.8444	5.3858
0.1314	6.3385	13000	0.1172	22.1439	5.1436
0.1381	6.8261	14000	0.1151	21.7782	5.0237
0.109	7.3136	15000	0.1202	22.0646	5.1681
0.1201	7.8013	16000	0.1097	21.4522	4.9377
0.1114	8.2887	17000	0.1222	21.7914	5.0063
0.1207	8.7764	18000	0.1208	21.5755	4.9456
0.1028	9.2638	19000	0.1126	21.4478	4.8753
0.1006	9.7515	20000	0.1140	21.1658	4.7846
0.093	10.2390	21000	0.1054	21.2054	4.7554
0.0865	10.7267	22000	0.1149	21.0468	4.7878
0.0678	11.2141	23000	0.1183	20.8618	4.7278
0.0742	11.7018	24000	0.1148	20.6283	4.6260
0.0762	12.1892	25000	0.1051	20.5269	4.5952
0.0743	12.6769	26000	0.1104	20.5402	4.5984
0.0729	13.1644	27000	0.1067	20.2362	4.5518
0.0722	13.6520	28000	0.1089	20.2009	4.5100
0.0638	14.1395	29000	0.1088	20.2273	4.5211
0.0717	14.6272	30000	0.1082	20.2758	4.5258
0.0592	15.1146	31000	0.1056	20.1524	4.4650
0.058	15.6023	32000	0.1058	19.9630	4.4508
0.0612	16.0897	33000	0.1099	19.8749	4.4098
0.0558	16.5774	34000	0.1105	19.8969	4.4138
0.0475	17.0649	35000	0.1115	19.9013	4.3830
0.0458	17.5525	36000	0.1032	19.8000	4.3727
0.0501	18.0400	37000	0.1078	19.6590	4.3262
0.043	18.5277	38000	0.1089	19.5621	4.3214
0.051	19.0151	39000	0.1084	19.5929	4.3120
0.0457	19.5028	40000	0.1078	19.5665	4.2733
0.0371	19.9905	41000	0.1078	19.3902	4.2457
0.0409	20.4779	42000	0.1083	19.4167	4.2276
0.034	20.9656	43000	0.1054	19.3902	4.2165
0.0365	21.4531	44000	0.1052	19.2052	4.1771
0.0553	21.9407	45000	0.1017	19.1523	4.1360
0.0408	22.4282	46000	0.1015	19.1876	4.1518
0.0466	22.9159	47000	0.1040	19.0906	4.1210
0.0454	23.4033	48000	0.1020	19.0774	4.1234
0.033	23.8910	49000	0.1064	19.1479	4.1479
0.0389	24.3784	50000	0.1043	19.0730	4.1297
0.0347	24.8661	51000	0.1056	19.0686	4.1297
0.0332	25.3536	52000	0.1026	18.9056	4.0697
0.0349	25.8413	53000	0.1036	18.9805	4.0824
0.0378	26.3287	54000	0.1007	18.9100	4.0563
0.0303	26.8164	55000	0.1016	18.8615	4.0477
0.0391	27.3038	56000	0.1015	18.8659	4.0492
0.0367	27.7915	57000	0.1016	18.8527	4.0508
0.0345	28.2790	58000	0.1019	18.8968	4.0658
0.0303	28.7666	59000	0.1016	18.8836	4.0595
0.0389	29.2541	60000	0.1016	18.9012	4.0626
0.0364	29.7418	61000	0.1016	18.9012	4.0682

Framework versions

Transformers 4.55.4
Pytorch 2.8.0+cu126
Datasets 4.0.0
Tokenizers 0.21.4

davidilag/wav2vec2-xls-r-300m-pt-1000h_faroese-best-faroese-100h-30-epochs_run7_2025-09-02

wav2vec2-xls-r-300m-pt-1000h_faroese-best-faroese-100h-30-epochs_run7_2025-09-02

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

davidilag/
wav2vec2-xls-r-300m-pt-1000h_faroese-best-faroese-100h-30-epochs_run7_2025-09-02