learning_rate: 0.0001
train_batch_size: 64
eval_batch_size: 128
seed: 42
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: 10000
num_epochs: 60
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
4.3828	0.9747	1000	3.9487	97.4093	99.1139
3.1191	1.9493	2000	3.1243	97.4358	99.1163
0.4799	2.9240	3000	0.3095	33.6741	8.7518
0.3163	3.8986	4000	0.1843	26.8714	6.5662
0.2328	4.8733	5000	0.1504	24.7037	5.8293
0.2321	5.8480	6000	0.1383	23.5670	5.5870
0.1761	6.8226	7000	0.1207	22.6462	5.2083
0.1498	7.7973	8000	0.1266	22.2673	5.1799
0.1242	8.7719	9000	0.1131	21.6372	4.9527
0.1522	9.7466	10000	0.1148	21.3200	4.9542
0.1128	10.7212	11000	0.1096	20.7693	4.7522
0.1349	11.6959	12000	0.1101	20.7516	4.7491
0.0915	12.6706	13000	0.1079	20.3111	4.5739
0.1276	13.6452	14000	0.1002	20.2802	4.5305
0.0847	14.6199	15000	0.0999	19.8044	4.3948
0.1094	15.5945	16000	0.1050	20.0423	4.4919
0.0793	16.5692	17000	0.1023	19.7559	4.4035
0.0988	17.5439	18000	0.0974	19.7559	4.3869
0.0601	18.5185	19000	0.1000	19.4651	4.3262
0.0973	19.4932	20000	0.1058	19.3109	4.2441
0.0579	20.4678	21000	0.1078	19.4563	4.3088
0.0801	21.4425	22000	0.1039	19.2889	4.2631
0.068	22.4172	23000	0.1012	19.1523	4.2473
0.0799	23.3918	24000	0.1159	19.5973	4.3167
0.0514	24.3665	25000	0.1014	18.9320	4.1542
0.0796	25.3411	26000	0.0992	18.9717	4.1660
0.0517	26.3158	27000	0.1019	18.9144	4.1068
0.0521	27.2904	28000	0.1022	18.7646	4.1068
0.0482	28.2651	29000	0.1058	18.9761	4.1487
0.0473	29.2398	30000	0.1042	18.7514	4.0697
0.0501	30.2144	31000	0.1071	18.9320	4.1021
0.0435	31.1891	32000	0.1007	18.6941	4.0264
0.0391	32.1637	33000	0.1088	18.5884	4.0232
0.0469	33.1384	34000	0.1072	18.4209	4.0035
0.0474	34.1131	35000	0.1094	18.5399	4.0129
0.0527	35.0877	36000	0.1070	18.4518	4.0019
0.0411	36.0624	37000	0.1043	18.1742	3.8938
0.0401	37.0370	38000	0.1072	18.2976	3.9056
0.0366	38.0117	39000	0.1077	18.2932	3.9325
0.0388	38.9864	40000	0.1066	18.2050	3.8678
0.032	39.9610	41000	0.1062	18.0597	3.8622
0.052	40.9357	42000	0.1047	18.1654	3.8575
0.0293	41.9103	43000	0.1084	17.8790	3.8086
0.032	42.8850	44000	0.1063	17.8658	3.8102
0.0237	43.8596	45000	0.1090	17.9319	3.8173
0.0386	44.8343	46000	0.1099	17.8526	3.8031
0.021	45.8090	47000	0.1120	17.8702	3.7928
0.0311	46.7836	48000	0.1102	17.8129	3.7762
0.0255	47.7583	49000	0.1130	17.8306	3.7762
0.0356	48.7329	50000	0.1096	17.8085	3.7873
0.0229	49.7076	51000	0.1104	17.8217	3.7810
0.0354	50.6823	52000	0.1116	17.6763	3.7352
0.0231	51.6569	53000	0.1113	17.7645	3.7581
0.0234	52.6316	54000	0.1098	17.6896	3.7407
0.0274	53.6062	55000	0.1090	17.6808	3.7336
0.0233	54.5809	56000	0.1098	17.7424	3.7470
0.0308	55.5556	57000	0.1095	17.7512	3.7486
0.02	56.5302	58000	0.1090	17.7336	3.7423
0.0167	57.5049	59000	0.1086	17.7380	3.7415
0.0274	58.4795	60000	0.1089	17.7336	3.7455
0.0313	59.4542	61000	0.1088	17.7336	3.7455

Framework versions

Transformers 4.56.1
Pytorch 2.8.0+cu126
Datasets 4.0.0
Tokenizers 0.22.0

davidilag/wav2vec2-xls-r-300m-cpt-1000h_faroese-cp_best-faroese-100h-60-epochs_run8_2025-09-24

wav2vec2-xls-r-300m-cpt-1000h_faroese-cp_best-faroese-100h-60-epochs_run8_2025-09-24

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-cp_best-faroese-100h-60-epochs_run8_2025-09-24