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
3.2941	0.4877	1000	3.2303	100.0	100.0
0.7524	0.9754	2000	0.4449	44.3054	11.9118
0.4103	1.4628	3000	0.2304	32.0263	8.0559
0.3529	1.9505	4000	0.2055	29.7705	7.3868
0.2663	2.4379	5000	0.1746	27.3560	6.6427
0.2557	2.9256	6000	0.1548	26.4220	6.3611
0.1854	3.4131	7000	0.1464	24.9769	5.9239
0.1918	3.9008	8000	0.1439	24.5054	5.8040
0.1525	4.3882	9000	0.1422	24.1309	5.6857
0.1655	4.8759	10000	0.1274	23.3511	5.4742
0.141	5.3633	11000	0.1294	23.2630	5.4363
0.1411	5.8510	12000	0.1239	22.9810	5.3227
0.119	6.3385	13000	0.1242	22.7211	5.2769
0.1268	6.8261	14000	0.1206	22.6153	5.2217
0.1067	7.3136	15000	0.1205	22.0293	5.1105
0.1098	7.8013	16000	0.1142	21.8135	4.9748
0.1179	8.2887	17000	0.1139	21.8972	4.9748
0.1054	8.7764	18000	0.1197	21.8090	4.9795
0.0962	9.2638	19000	0.1123	21.6328	4.9156
0.0937	9.7515	20000	0.1253	21.7826	4.9661
0.0837	10.2390	21000	0.1164	21.2143	4.7854
0.0857	10.7267	22000	0.1161	21.2495	4.7704
0.0728	11.2141	23000	0.1145	21.2054	4.8043
0.075	11.7018	24000	0.1181	21.0909	4.7515
0.0782	12.1892	25000	0.1093	20.8265	4.6710
0.0694	12.6769	26000	0.1176	20.6503	4.6244
0.0702	13.1644	27000	0.1104	20.6811	4.6268
0.0671	13.6520	28000	0.1088	20.4653	4.5668
0.0625	14.1395	29000	0.1134	20.5049	4.5747
0.066	14.6272	30000	0.1098	20.3331	4.5179
0.0582	15.1146	31000	0.1109	20.2362	4.5045
0.0604	15.6023	32000	0.1112	20.3199	4.5329
0.0628	16.0897	33000	0.1104	20.2802	4.4982
0.0575	16.5774	34000	0.1127	20.4256	4.5029
0.0498	17.0649	35000	0.1118	20.0775	4.4501
0.0485	17.5525	36000	0.1113	20.1348	4.4445
0.0497	18.0400	37000	0.1108	20.0115	4.4232
0.0428	18.5277	38000	0.1149	20.0026	4.3909
0.0485	19.0151	39000	0.1100	19.9013	4.3625
0.0457	19.5028	40000	0.1092	19.6722	4.3207
0.0407	19.9905	41000	0.1070	19.4960	4.2544
0.0452	20.4779	42000	0.1086	19.6105	4.2867
0.0357	20.9656	43000	0.1103	19.5268	4.2567
0.0339	21.4531	44000	0.1082	19.5797	4.2575
0.053	21.9407	45000	0.1091	19.5488	4.2386
0.0404	22.4282	46000	0.1123	19.4343	4.2410
0.0472	22.9159	47000	0.1101	19.4739	4.2197
0.0445	23.4033	48000	0.1056	19.2801	4.1597
0.0354	23.8910	49000	0.1103	19.4739	4.2165
0.0394	24.3784	50000	0.1096	19.3990	4.2039
0.0389	24.8661	51000	0.1101	19.3814	4.1842
0.035	25.3536	52000	0.1102	19.3153	4.1747
0.0322	25.8413	53000	0.1084	19.2448	4.1455
0.0373	26.3287	54000	0.1074	19.2184	4.1329
0.0309	26.8164	55000	0.1088	19.1876	4.1360
0.0346	27.3038	56000	0.1096	19.1787	4.1210
0.0399	27.7915	57000	0.1081	19.1611	4.1266
0.0386	28.2790	58000	0.1084	19.1743	4.1258
0.0316	28.7666	59000	0.1086	19.1832	4.1242
0.0406	29.2541	60000	0.1082	19.2052	4.1234
0.0389	29.7418	61000	0.1082	19.2052	4.1250

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_67h_faroese-best-faroese-100h-30-epochs_run6_2025-08-29

wav2vec2-xls-r-300m-pt-1000h_67h_faroese-best-faroese-100h-30-epochs_run6_2025-08-29

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

davidilag/
wav2vec2-xls-r-300m-pt-1000h_67h_faroese-best-faroese-100h-30-epochs_run6_2025-08-29