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
3.3093	0.4877	1000	3.2256	100.0	100.0
1.0068	0.9754	2000	0.6273	52.5444	14.7696
0.4698	1.4628	3000	0.2536	33.5639	8.6823
0.4211	1.9505	4000	0.2233	30.5591	7.6953
0.3471	2.4379	5000	0.1800	28.6822	7.0191
0.3156	2.9256	6000	0.1648	27.0432	6.6033
0.2488	3.4131	7000	0.1592	25.6554	6.2127
0.239	3.9008	8000	0.1528	25.3514	6.1015
0.2144	4.3882	9000	0.1420	24.5759	5.8529
0.2255	4.8759	10000	0.1408	24.3645	5.8009
0.189	5.3633	11000	0.1417	24.3468	5.7953
0.1997	5.8510	12000	0.1287	24.0428	5.7188
0.1658	6.3385	13000	0.1286	23.8622	5.6186
0.1712	6.8261	14000	0.1273	22.9546	5.3503
0.1393	7.3136	15000	0.1210	22.8048	5.3558
0.1458	7.8013	16000	0.1226	22.5228	5.2399
0.1459	8.2887	17000	0.1176	22.1219	5.1018
0.1435	8.7764	18000	0.1209	22.1747	5.1270
0.12	9.2638	19000	0.1132	21.8663	5.0032
0.1279	9.7515	20000	0.1173	21.9456	5.0623
0.1167	10.2390	21000	0.1165	21.8443	5.0205
0.1078	10.7267	22000	0.1127	21.5447	4.9172
0.0968	11.2141	23000	0.1128	21.2980	4.7996
0.0933	11.7018	24000	0.1097	21.3024	4.8611
0.0975	12.1892	25000	0.1031	21.0601	4.7901
0.0897	12.6769	26000	0.1085	20.8354	4.7096
0.0901	13.1644	27000	0.1089	20.9852	4.7215
0.0834	13.6520	28000	0.1111	20.7076	4.6284
0.0765	14.1395	29000	0.1077	20.7472	4.6039
0.0877	14.6272	30000	0.1070	20.6503	4.6031
0.0745	15.1146	31000	0.1096	20.7208	4.6386
0.0737	15.6023	32000	0.1051	20.5622	4.5779
0.0801	16.0897	33000	0.1061	20.4829	4.5597
0.0706	16.5774	34000	0.1059	20.3419	4.5739
0.0628	17.0649	35000	0.1059	20.1921	4.4572
0.055	17.5525	36000	0.1077	20.1568	4.4635
0.0642	18.0400	37000	0.1018	19.8749	4.3869
0.0542	18.5277	38000	0.1084	19.7956	4.3712
0.0587	19.0151	39000	0.1035	19.8837	4.3854
0.0565	19.5028	40000	0.1054	19.6766	4.3570
0.0481	19.9905	41000	0.1043	19.5753	4.2970
0.0539	20.4779	42000	0.1002	19.6370	4.3317
0.0425	20.9656	43000	0.1027	19.5488	4.2828
0.0423	21.4531	44000	0.1029	19.5136	4.2844
0.0637	21.9407	45000	0.1025	19.3109	4.2165
0.0544	22.4282	46000	0.1004	19.3241	4.2244
0.0561	22.9159	47000	0.0999	19.3109	4.2086
0.0534	23.4033	48000	0.0967	19.2184	4.1731
0.0403	23.8910	49000	0.0992	19.2052	4.1857
0.0449	24.3784	50000	0.0978	19.1523	4.1581
0.0433	24.8661	51000	0.0983	19.1523	4.1321
0.0425	25.3536	52000	0.0993	19.1567	4.1463
0.0415	25.8413	53000	0.0999	19.0730	4.1273
0.0436	26.3287	54000	0.1003	19.1215	4.1463
0.0378	26.8164	55000	0.0999	19.1391	4.1297
0.0426	27.3038	56000	0.1011	19.1347	4.1266
0.0481	27.7915	57000	0.0994	19.1215	4.1218
0.0466	28.2790	58000	0.1003	19.1215	4.1226
0.0349	28.7666	59000	0.1001	19.0906	4.1195
0.0492	29.2541	60000	0.0998	19.0906	4.1187
0.0503	29.7418	61000	0.0997	19.0862	4.1179

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-last-faroese-100h-30-epochs_run2_2025-08-25

wav2vec2-xls-r-300m-pre_trained-1000h_faroese-last-faroese-100h-30-epochs_2025-08-25

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-last-faroese-100h-30-epochs_run2_2025-08-25