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
3.3239	0.4877	1000	3.2507	100.0	100.0
0.7706	0.9754	2000	0.4437	39.6837	10.8411
0.3957	1.4628	3000	0.2259	29.7176	7.6803
0.3567	1.9505	4000	0.1970	27.7702	6.9781
0.2861	2.4379	5000	0.1731	26.3515	6.4344
0.2722	2.9256	6000	0.1523	25.1795	6.1370
0.1925	3.4131	7000	0.1478	24.6685	5.9003
0.2167	3.9008	8000	0.1411	23.9679	5.7835
0.1615	4.3882	9000	0.1409	23.5317	5.6288
0.1771	4.8759	10000	0.1380	23.0823	5.4624
0.1445	5.3633	11000	0.1367	23.0779	5.4103
0.1572	5.8510	12000	0.1277	23.1176	5.4387
0.1335	6.3385	13000	0.1157	22.4082	5.1823
0.1334	6.8261	14000	0.1218	21.9853	5.0647
0.1176	7.3136	15000	0.1142	22.1351	5.1105
0.1188	7.8013	16000	0.1126	21.6152	4.9787
0.119	8.2887	17000	0.1217	21.5359	5.0071
0.1147	8.7764	18000	0.1166	21.4257	4.9235
0.0991	9.2638	19000	0.1188	21.2275	4.8824
0.1068	9.7515	20000	0.1089	21.3597	4.8525
0.092	10.2390	21000	0.1125	20.9543	4.7617
0.0917	10.7267	22000	0.1109	20.9940	4.7909
0.0723	11.2141	23000	0.1114	20.8926	4.7617
0.0801	11.7018	24000	0.1091	20.8574	4.7152
0.0765	12.1892	25000	0.1045	20.4564	4.5858
0.0769	12.6769	26000	0.1094	20.7649	4.6741
0.0717	13.1644	27000	0.1041	20.4168	4.5439
0.0722	13.6520	28000	0.1109	20.5005	4.6102
0.0654	14.1395	29000	0.1074	20.2890	4.5345
0.0733	14.6272	30000	0.1035	20.0996	4.5132
0.0621	15.1146	31000	0.1046	20.0731	4.4619
0.0601	15.6023	32000	0.1015	20.0291	4.4295
0.0671	16.0897	33000	0.1007	20.0291	4.4217
0.0578	16.5774	34000	0.1037	19.7471	4.3735
0.0513	17.0649	35000	0.1034	19.8440	4.3790
0.0485	17.5525	36000	0.1000	19.6149	4.3167
0.0524	18.0400	37000	0.1033	19.6634	4.3333
0.0446	18.5277	38000	0.1030	19.4563	4.2378
0.0534	19.0151	39000	0.1021	19.5929	4.2915
0.0449	19.5028	40000	0.1045	19.4828	4.2623
0.0394	19.9905	41000	0.1006	19.4475	4.2473
0.0431	20.4779	42000	0.1010	19.4872	4.2386
0.0361	20.9656	43000	0.1007	19.4343	4.2299
0.0348	21.4531	44000	0.0997	19.4079	4.2070
0.0534	21.9407	45000	0.0985	19.3021	4.1723
0.0406	22.4282	46000	0.0994	19.2404	4.1526
0.0447	22.9159	47000	0.0997	19.0774	4.1195
0.0486	23.4033	48000	0.0961	19.1567	4.1273
0.0342	23.8910	49000	0.0980	19.0818	4.1187
0.042	24.3784	50000	0.0984	19.0245	4.0942
0.0351	24.8661	51000	0.0989	18.9496	4.0674
0.0351	25.3536	52000	0.0998	18.9981	4.0792
0.0352	25.8413	53000	0.0992	18.9893	4.0619
0.0381	26.3287	54000	0.0981	18.9100	4.0540
0.0306	26.8164	55000	0.0989	18.9496	4.0548
0.0352	27.3038	56000	0.0991	18.9452	4.0540
0.0392	27.7915	57000	0.0980	18.9320	4.0437
0.0355	28.2790	58000	0.0988	18.9805	4.0508
0.0324	28.7666	59000	0.0984	18.9364	4.0461
0.0429	29.2541	60000	0.0979	18.9276	4.0413
0.039	29.7418	61000	0.0979	18.9276	4.0413

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-cp12-faroese-100h-30-epochs_run7_2025-09-02

wav2vec2-xls-r-300m-pt-1000h_faroese-cp12-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-cp12-faroese-100h-30-epochs_run7_2025-09-02