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
3.3375	0.4877	1000	3.2726	100.0	98.0559
0.892	0.9754	2000	0.5545	47.3455	13.2997
0.4238	1.4628	3000	0.2568	32.5021	8.4164
0.3759	1.9505	4000	0.2101	29.3034	7.4373
0.3082	2.4379	5000	0.1826	27.9552	6.8913
0.2799	2.9256	6000	0.1591	26.0255	6.3453
0.2067	3.4131	7000	0.1525	25.1179	6.0447
0.2151	3.9008	8000	0.1527	24.8315	5.9934
0.1632	4.3882	9000	0.1444	24.3600	5.8087
0.1803	4.8759	10000	0.1325	23.4613	5.5421
0.1574	5.3633	11000	0.1509	23.5934	5.6107
0.1603	5.8510	12000	0.1270	23.4524	5.5302
0.1345	6.3385	13000	0.1305	23.1528	5.4734
0.1388	6.8261	14000	0.1231	22.9942	5.3219
0.117	7.3136	15000	0.1250	22.7475	5.3243
0.1307	7.8013	16000	0.1202	22.6990	5.1625
0.121	8.2887	17000	0.1195	22.3378	5.1460
0.1217	8.7764	18000	0.1239	22.0954	5.1475
0.1019	9.2638	19000	0.1199	21.8575	5.0537
0.1043	9.7515	20000	0.1229	21.8355	5.0032
0.0939	10.2390	21000	0.1180	21.6857	4.9716
0.1	10.7267	22000	0.1161	21.5711	4.9740
0.0794	11.2141	23000	0.1167	21.3112	4.8430
0.0835	11.7018	24000	0.1171	21.3376	4.8611
0.0817	12.1892	25000	0.1115	21.0028	4.7846
0.0787	12.6769	26000	0.1165	21.2363	4.8051
0.0742	13.1644	27000	0.1200	20.9719	4.7633
0.0752	13.6520	28000	0.1139	20.9058	4.7309
0.0695	14.1395	29000	0.1157	20.9235	4.7380
0.0801	14.6272	30000	0.1124	20.8926	4.7215
0.0687	15.1146	31000	0.1156	20.6459	4.6315
0.062	15.6023	32000	0.1166	20.7560	4.6323
0.0721	16.0897	33000	0.1105	20.4653	4.5487
0.0648	16.5774	34000	0.1118	20.4036	4.5424
0.0537	17.0649	35000	0.1086	20.2890	4.4950
0.0518	17.5525	36000	0.1108	20.1613	4.5061
0.0591	18.0400	37000	0.1116	19.9674	4.4485
0.0499	18.5277	38000	0.1161	19.9454	4.4430
0.0518	19.0151	39000	0.1191	20.1921	4.5203
0.0491	19.5028	40000	0.1142	20.0379	4.4461
0.0447	19.9905	41000	0.1133	19.8220	4.3830
0.0471	20.4779	42000	0.1124	19.9454	4.3877
0.0415	20.9656	43000	0.1101	19.7647	4.3380
0.0394	21.4531	44000	0.1101	19.7471	4.3420
0.0558	21.9407	45000	0.1116	19.6986	4.3167
0.0434	22.4282	46000	0.1129	19.6281	4.3396
0.0493	22.9159	47000	0.1122	19.6546	4.2938
0.05	23.4033	48000	0.1124	19.5092	4.2773
0.0358	23.8910	49000	0.1136	19.6193	4.2717
0.0445	24.3784	50000	0.1120	19.5268	4.2670
0.0422	24.8661	51000	0.1095	19.4828	4.2512
0.0385	25.3536	52000	0.1096	19.4079	4.2220
0.0385	25.8413	53000	0.1106	19.3726	4.2118
0.0432	26.3287	54000	0.1096	19.3374	4.2039
0.0337	26.8164	55000	0.1101	19.3638	4.2110
0.0416	27.3038	56000	0.1097	19.3285	4.2094
0.0453	27.7915	57000	0.1094	19.3418	4.2134
0.0381	28.2790	58000	0.1100	19.3197	4.2126
0.032	28.7666	59000	0.1098	19.2713	4.2007
0.0486	29.2541	60000	0.1097	19.2845	4.2007
0.0424	29.7418	61000	0.1097	19.2977	4.2015

Framework versions

Transformers 4.55.2
Pytorch 2.8.0+cu126
Datasets 4.0.0
Tokenizers 0.21.4

davidilag/wav2vec2-xls-r-300m-pt-1000h_faroese-best-faroese-100h-30-epochs_run4_2025-08-26

wav2vec2-xls-r-300m-pt-1000h_67h_faroese-best-faroese-100h-30-epochs_run4_2025-08-26

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-best-faroese-100h-30-epochs_run4_2025-08-26