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
3.4031	0.4877	1000	3.3166	100.0	99.2560
0.841	0.9754	2000	0.4918	44.2261	12.1777
0.4075	1.4628	3000	0.2292	30.5988	7.8207
0.3696	1.9505	4000	0.1940	28.3121	7.0909
0.2937	2.4379	5000	0.1744	26.8934	6.6001
0.2724	2.9256	6000	0.1554	25.8316	6.2577
0.2018	3.4131	7000	0.1500	24.8491	5.9902
0.221	3.9008	8000	0.1398	24.2455	5.7646
0.1636	4.3882	9000	0.1468	24.2631	5.7480
0.1819	4.8759	10000	0.1354	23.4833	5.5492
0.1526	5.3633	11000	0.1384	23.1484	5.4923
0.1502	5.8510	12000	0.1291	23.1793	5.4647
0.1269	6.3385	13000	0.1248	22.4964	5.3203
0.137	6.8261	14000	0.1287	22.2408	5.1894
0.1181	7.3136	15000	0.1166	22.1880	5.1965
0.1275	7.8013	16000	0.1181	22.1439	5.1428
0.1224	8.2887	17000	0.1210	21.7738	5.0142
0.1175	8.7764	18000	0.1175	21.8399	4.9755
0.1035	9.2638	19000	0.1142	21.6284	4.9897
0.1033	9.7515	20000	0.1205	21.5447	4.9755
0.0908	10.2390	21000	0.1187	21.4434	4.8951
0.0916	10.7267	22000	0.1207	21.5095	4.9290
0.0747	11.2141	23000	0.1191	21.2539	4.8020
0.0809	11.7018	24000	0.1132	20.9940	4.7593
0.0812	12.1892	25000	0.1138	21.0512	4.7885
0.0745	12.6769	26000	0.1189	21.1041	4.7215
0.0739	13.1644	27000	0.1161	20.7164	4.6544
0.0706	13.6520	28000	0.1085	20.6856	4.6505
0.0712	14.1395	29000	0.1110	20.4961	4.5724
0.0788	14.6272	30000	0.1087	20.4300	4.5329
0.0607	15.1146	31000	0.1096	20.3815	4.5526
0.0596	15.6023	32000	0.1111	20.4520	4.5700
0.0686	16.0897	33000	0.1076	20.1745	4.5132
0.0669	16.5774	34000	0.1027	20.0555	4.4382
0.0521	17.0649	35000	0.1029	20.0731	4.4319
0.05	17.5525	36000	0.1025	19.8969	4.4177
0.0546	18.0400	37000	0.1021	19.7779	4.3648
0.0472	18.5277	38000	0.1080	19.8396	4.3932
0.0517	19.0151	39000	0.1023	19.7383	4.3412
0.0472	19.5028	40000	0.1064	19.7295	4.3459
0.0382	19.9905	41000	0.1113	19.4475	4.2970
0.046	20.4779	42000	0.1045	19.5312	4.3128
0.0365	20.9656	43000	0.1063	19.4607	4.2686
0.0345	21.4531	44000	0.1037	19.4783	4.2567
0.0584	21.9407	45000	0.0992	19.2801	4.2078
0.045	22.4282	46000	0.1021	19.1876	4.1786
0.0456	22.9159	47000	0.1043	19.1479	4.1518
0.0485	23.4033	48000	0.1018	19.0245	4.1124
0.0351	23.8910	49000	0.1041	19.1303	4.1755
0.0421	24.3784	50000	0.1016	18.9893	4.1234
0.0358	24.8661	51000	0.1056	18.9452	4.1013
0.0344	25.3536	52000	0.1048	18.9717	4.1124
0.0363	25.8413	53000	0.1034	18.9496	4.1060
0.0383	26.3287	54000	0.1047	18.8968	4.0792
0.035	26.8164	55000	0.1035	18.8880	4.0753
0.0387	27.3038	56000	0.1046	18.8615	4.0642
0.0423	27.7915	57000	0.1040	18.9012	4.0697
0.0393	28.2790	58000	0.1046	18.8615	4.0697
0.03	28.7666	59000	0.1041	18.8703	4.0713
0.0438	29.2541	60000	0.1040	18.8880	4.0690
0.0398	29.7418	61000	0.1039	18.8880	4.0690

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-cp10-faroese-100h-30-epochs_run7_2025-09-02

wav2vec2-xls-r-300m-pt-1000h_faroese-cp10-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-cp10-faroese-100h-30-epochs_run7_2025-09-02