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
3.317	0.4877	1000	3.2380	100.0	98.9033
1.102	0.9754	2000	0.7354	61.3297	17.9967
0.5097	1.4628	3000	0.2764	34.9385	9.1195
0.454	1.9505	4000	0.2514	32.5726	8.3375
0.3519	2.4379	5000	0.2089	30.0128	7.5951
0.3298	2.9256	6000	0.1867	28.6910	7.1714
0.2579	3.4131	7000	0.1628	26.7568	6.5481
0.278	3.9008	8000	0.1645	26.3427	6.4999
0.2188	4.3882	9000	0.1538	25.7435	6.1859
0.2359	4.8759	10000	0.1443	25.1575	6.0218
0.1958	5.3633	11000	0.1424	24.7654	5.8711
0.2106	5.8510	12000	0.1299	24.3865	5.7314
0.1733	6.3385	13000	0.1352	24.0428	5.7401
0.183	6.8261	14000	0.1222	23.3555	5.4245
0.1511	7.3136	15000	0.1259	23.1396	5.4292
0.1618	7.8013	16000	0.1237	22.8576	5.3464
0.1522	8.2887	17000	0.1234	22.8532	5.2809
0.1601	8.7764	18000	0.1190	22.9017	5.3196
0.1282	9.2638	19000	0.1172	22.2937	5.1570
0.1375	9.7515	20000	0.1221	22.3950	5.1602
0.1253	10.2390	21000	0.1126	22.0602	5.0458
0.1133	10.7267	22000	0.1203	22.1395	5.0544
0.1007	11.2141	23000	0.1165	21.7121	4.9621
0.104	11.7018	24000	0.1149	21.7033	4.9669
0.0999	12.1892	25000	0.1079	21.1394	4.8256
0.0975	12.6769	26000	0.1135	21.4037	4.8525
0.1021	13.1644	27000	0.1136	21.1878	4.7704
0.0889	13.6520	28000	0.1104	21.2143	4.7420
0.0858	14.1395	29000	0.1056	21.0556	4.7096
0.0916	14.6272	30000	0.1114	20.7472	4.6686
0.0817	15.1146	31000	0.1114	21.0160	4.6978
0.0747	15.6023	32000	0.1090	20.6988	4.6023
0.0837	16.0897	33000	0.1025	20.6371	4.5866
0.0748	16.5774	34000	0.1087	20.4697	4.5597
0.067	17.0649	35000	0.1022	20.3243	4.4769
0.0608	17.5525	36000	0.1067	20.3771	4.5092
0.0634	18.0400	37000	0.1077	20.1304	4.4792
0.0598	18.5277	38000	0.1102	20.0775	4.4382
0.0626	19.0151	39000	0.1054	20.1701	4.4572
0.0565	19.5028	40000	0.1055	20.0599	4.4248
0.054	19.9905	41000	0.1043	19.9277	4.3790
0.0526	20.4779	42000	0.1032	19.8176	4.3396
0.0452	20.9656	43000	0.1024	19.6810	4.2883
0.0444	21.4531	44000	0.1001	19.6898	4.2915
0.0709	21.9407	45000	0.0999	19.6458	4.2678
0.0545	22.4282	46000	0.0978	19.6149	4.2631
0.0619	22.9159	47000	0.1014	19.6237	4.2662
0.0553	23.4033	48000	0.0971	19.4079	4.2086
0.0441	23.8910	49000	0.1015	19.5577	4.2260
0.0475	24.3784	50000	0.0981	19.3770	4.1889
0.046	24.8661	51000	0.0984	19.4475	4.2055
0.0455	25.3536	52000	0.0984	19.3418	4.1692
0.0413	25.8413	53000	0.0977	19.3374	4.1684
0.0469	26.3287	54000	0.0985	19.2360	4.1487
0.0426	26.8164	55000	0.0988	19.2889	4.1479
0.0471	27.3038	56000	0.0984	19.2536	4.1550
0.053	27.7915	57000	0.0979	19.2757	4.1479
0.0498	28.2790	58000	0.0984	19.2536	4.1431
0.0352	28.7666	59000	0.0981	19.2492	4.1392
0.0548	29.2541	60000	0.0979	19.2404	4.1368
0.0507	29.7418	61000	0.0978	19.2492	4.1376

Framework versions

Transformers 4.55.2
Pytorch 2.8.0+cu126
Datasets 4.0.0
Tokenizers 0.21.4

davidilag/wav2vec2-xls-r-300m-cpt-1000h_faroese-5_epochs-faroese-100h-30-epochs_run2_2025-08-22

wav2vec2-xls-r-300m-pre_trained-1000h_faroese-5_epochs-faroese-100h-30-epochs_2025-08-22

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

davidilag/
wav2vec2-xls-r-300m-cpt-1000h_faroese-5_epochs-faroese-100h-30-epochs_run2_2025-08-22