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
3.3012	0.4877	1000	3.2436	100.0	97.7608
0.8791	0.9754	2000	0.5321	46.1647	12.8412
0.4303	1.4628	3000	0.2401	31.0482	7.9588
0.3871	1.9505	4000	0.2027	29.2197	7.2621
0.3071	2.4379	5000	0.1781	27.4265	6.7611
0.2748	2.9256	6000	0.1687	26.2898	6.4565
0.205	3.4131	7000	0.1483	24.5098	5.8790
0.2241	3.9008	8000	0.1432	24.3997	5.8616
0.1719	4.3882	9000	0.1419	23.9415	5.7148
0.1842	4.8759	10000	0.1306	23.8093	5.6241
0.1545	5.3633	11000	0.1353	23.2498	5.5105
0.156	5.8510	12000	0.1338	23.2938	5.4545
0.1372	6.3385	13000	0.1217	22.9149	5.3551
0.1433	6.8261	14000	0.1379	22.5669	5.2683
0.1161	7.3136	15000	0.1277	22.3466	5.2391
0.1254	7.8013	16000	0.1201	22.0073	5.0726
0.1302	8.2887	17000	0.1158	21.7782	5.0103
0.1271	8.7764	18000	0.1128	21.5888	4.9408
0.1031	9.2638	19000	0.1134	21.6504	4.9740
0.1031	9.7515	20000	0.1232	21.3949	4.9061
0.0967	10.2390	21000	0.1134	21.4125	4.8698
0.0949	10.7267	22000	0.1135	21.5844	4.8943
0.0796	11.2141	23000	0.1203	21.3464	4.8872
0.08	11.7018	24000	0.1146	21.2275	4.8430
0.0791	12.1892	25000	0.1114	21.1217	4.8020
0.0783	12.6769	26000	0.1177	20.9499	4.7199
0.0766	13.1644	27000	0.1172	20.7605	4.6757
0.0677	13.6520	28000	0.1176	20.5974	4.6639
0.0672	14.1395	29000	0.1182	20.6415	4.6371
0.0739	14.6272	30000	0.1152	20.5754	4.5850
0.0647	15.1146	31000	0.1161	20.5005	4.5992
0.0608	15.6023	32000	0.1102	20.4212	4.5731
0.0692	16.0897	33000	0.1106	20.2802	4.5526
0.0642	16.5774	34000	0.1159	20.1568	4.5108
0.0541	17.0649	35000	0.1122	20.1348	4.4658
0.0488	17.5525	36000	0.1081	19.9850	4.4193
0.0539	18.0400	37000	0.1138	19.9101	4.4185
0.0478	18.5277	38000	0.1079	19.8000	4.3917
0.0543	19.0151	39000	0.1135	19.8176	4.3956
0.052	19.5028	40000	0.1120	19.7427	4.3830
0.0437	19.9905	41000	0.1103	19.6590	4.3293
0.0462	20.4779	42000	0.1132	19.6149	4.3151
0.0374	20.9656	43000	0.1068	19.6722	4.2891
0.0364	21.4531	44000	0.1087	19.5929	4.3072
0.0595	21.9407	45000	0.1040	19.3682	4.2181
0.0449	22.4282	46000	0.1076	19.4343	4.2441
0.0474	22.9159	47000	0.1080	19.4079	4.2425
0.0456	23.4033	48000	0.1043	19.3241	4.2228
0.0368	23.8910	49000	0.1047	19.1743	4.1849
0.0432	24.3784	50000	0.1035	19.2669	4.1857
0.0379	24.8661	51000	0.1071	19.2184	4.1794
0.0371	25.3536	52000	0.1050	19.2008	4.1802
0.0348	25.8413	53000	0.1077	19.1787	4.1763
0.0427	26.3287	54000	0.1063	19.2096	4.1897
0.035	26.8164	55000	0.1066	19.1964	4.1889
0.0395	27.3038	56000	0.1064	19.1259	4.1550
0.04	27.7915	57000	0.1066	19.1479	4.1510
0.0406	28.2790	58000	0.1070	19.1655	4.1613
0.0308	28.7666	59000	0.1064	19.1611	4.1573
0.0427	29.2541	60000	0.1063	19.1743	4.1621
0.0387	29.7418	61000	0.1061	19.1699	4.1613

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-cp8-faroese-100h-30-epochs_run7_2025-09-03

wav2vec2-xls-r-300m-pt-1000h_faroese-cp8-faroese-100h-30-epochs_run7_2025-09-03

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-cp8-faroese-100h-30-epochs_run7_2025-09-03