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 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
3.3186	0.4877	1000	3.2374	100.0	99.9992
1.261	0.9754	2000	0.8817	71.1239	21.7903
0.6199	1.4628	3000	0.3426	41.8690	11.1985
0.4976	1.9505	4000	0.2694	36.4365	9.4027
0.3769	2.4379	5000	0.2169	31.5108	7.8902
0.3502	2.9256	6000	0.1904	29.9731	7.3221
0.2701	3.4131	7000	0.1674	27.8495	6.7232
0.275	3.9008	8000	0.1534	26.8802	6.4518
0.2219	4.3882	9000	0.1527	26.5850	6.3855
0.2399	4.8759	10000	0.1408	25.2985	6.0305
0.1945	5.3633	11000	0.1464	25.4747	6.0305
0.1953	5.8510	12000	0.1311	24.7654	5.7724
0.1626	6.3385	13000	0.1321	24.3292	5.7212
0.1752	6.8261	14000	0.1313	23.7829	5.5594
0.1493	7.3136	15000	0.1320	23.4260	5.4616
0.1526	7.8013	16000	0.1254	22.9898	5.3606
0.14	8.2887	17000	0.1223	22.8532	5.2351
0.1454	8.7764	18000	0.1251	22.4171	5.1531
0.1218	9.2638	19000	0.1152	22.2012	5.0789
0.1239	9.7515	20000	0.1223	22.0337	5.0442
0.1133	10.2390	21000	0.1155	22.1835	5.0331
0.1073	10.7267	22000	0.1159	22.2496	5.0544
0.0949	11.2141	23000	0.1169	21.7253	4.9243
0.093	11.7018	24000	0.1157	21.7430	4.9456
0.0963	12.1892	25000	0.1128	21.6504	4.9337
0.0913	12.6769	26000	0.1122	21.4786	4.8698
0.0853	13.1644	27000	0.1133	21.2407	4.8162
0.0824	13.6520	28000	0.1084	21.1746	4.7467
0.074	14.1395	29000	0.1127	21.0160	4.7089
0.0888	14.6272	30000	0.1070	21.0248	4.7057
0.0757	15.1146	31000	0.1128	20.8706	4.6702
0.0699	15.6023	32000	0.1045	20.8398	4.6244
0.0728	16.0897	33000	0.1077	20.6988	4.5929
0.0678	16.5774	34000	0.1064	20.5137	4.5242
0.0592	17.0649	35000	0.1040	20.4476	4.5021
0.0545	17.5525	36000	0.1100	20.4476	4.5084
0.0652	18.0400	37000	0.1025	20.2978	4.4437
0.0538	18.5277	38000	0.1044	20.0423	4.3901
0.0578	19.0151	39000	0.1043	19.9674	4.3767
0.0537	19.5028	40000	0.1060	19.9189	4.3506
0.0493	19.9905	41000	0.1017	19.8837	4.3199
0.0544	20.4779	42000	0.0990	19.8176	4.3017
0.0429	20.9656	43000	0.1009	19.7163	4.2694
0.0389	21.4531	44000	0.0994	19.6678	4.2354
0.0651	21.9407	45000	0.0988	19.6986	4.2394
0.0544	22.4282	46000	0.0978	19.5797	4.1991
0.0577	22.9159	47000	0.0993	19.6193	4.2181
0.0555	23.4033	48000	0.0971	19.6590	4.1984
0.0418	23.8910	49000	0.0997	19.4034	4.1739
0.0524	24.3784	50000	0.0976	19.4739	4.1660
0.0424	24.8661	51000	0.0980	19.4079	4.1526
0.0412	25.3536	52000	0.0992	19.3021	4.1400
0.0394	25.8413	53000	0.0992	19.3506	4.1621
0.0459	26.3287	54000	0.0978	19.3021	4.1337
0.0391	26.8164	55000	0.0987	19.2536	4.1179
0.0434	27.3038	56000	0.0985	19.2492	4.1037
0.0494	27.7915	57000	0.0982	19.2228	4.0982
0.0417	28.2790	58000	0.0997	19.2360	4.1084
0.0417	28.7666	59000	0.0991	19.2360	4.1076
0.0509	29.2541	60000	0.0990	19.2492	4.1092
0.0498	29.7418	61000	0.0990	19.2404	4.1084

Framework versions

Transformers 4.53.1
Pytorch 2.6.0+cu124
Datasets 2.14.4
Tokenizers 0.21.2

davidilag/wav2vec2-xls-r-300m-converted-faroese-100h-30-epochs_2025-07-10_v2

wav2vec2-xls-r-300m-converted-faroese-100h-30-epochs_2025-07-10_v2

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

davidilag/
wav2vec2-xls-r-300m-converted-faroese-100h-30-epochs_2025-07-10_v2