wav2vec2-xls-r-1b-scandinavian-E2-100h-30-epochs-20250123

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1348
• Wer: 21.6270
• Cer: 4.5904

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• 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: 6000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.0569	0.5831	1000	0.5118	63.8799	16.0725
0.4643	1.1662	2000	0.2647	38.1428	9.2101
0.3797	1.7493	3000	0.2158	33.0390	7.8553
0.3558	2.3324	4000	0.1996	31.8993	7.4365
0.3784	2.9155	5000	0.2039	31.7718	7.5216
0.4332	3.4985	6000	0.2181	33.0704	7.8748
0.3112	4.0816	7000	0.2070	32.4627	7.6624
0.3189	4.6647	8000	0.1970	31.6000	7.4421
0.2768	5.2478	9000	0.1998	31.2380	7.4350
0.2924	5.8309	10000	0.1871	30.2479	7.0815
0.2971	6.4140	11000	0.1874	30.3957	7.2106
0.3326	6.9971	12000	0.1860	29.7972	7.0494
0.2833	7.5802	13000	0.1908	30.0964	7.1554
0.2283	8.1633	14000	0.1775	29.1211	6.8593
0.2145	8.7464	15000	0.1768	28.5836	6.6787
0.2556	9.3294	16000	0.1785	28.7831	6.7674
0.2567	9.9125	17000	0.1738	28.3009	6.5702
0.2418	10.4956	18000	0.1719	28.5466	6.6845
0.1733	11.0787	19000	0.1619	27.0467	6.2287
0.1734	11.6618	20000	0.1624	26.6052	6.1501
0.2026	12.2449	21000	0.1614	26.8417	6.1794
0.1898	12.8280	22000	0.1541	26.6902	6.1292
0.2149	13.4111	23000	0.1570	26.3780	6.0525
0.1751	13.9942	24000	0.1494	25.8756	5.8931
0.1488	14.5773	25000	0.1468	25.5819	5.7224
0.1283	15.1603	26000	0.1477	25.3380	5.6983
0.1363	15.7434	27000	0.1463	24.8430	5.5717
0.1324	16.3265	28000	0.1453	24.9095	5.5905
0.143	16.9096	29000	0.1434	24.6915	5.5245
0.1241	17.4927	30000	0.1436	24.1965	5.3843
0.0976	18.0758	31000	0.1475	24.1429	5.3458
0.0945	18.6589	32000	0.1405	23.4871	5.1769
0.1085	19.2420	33000	0.1393	23.6515	5.1972
0.1165	19.8251	34000	0.1350	23.2599	5.1165
0.1195	20.4082	35000	0.1435	23.1196	5.0626
0.101	20.9913	36000	0.1366	22.9219	4.9904
0.0764	21.5743	37000	0.1359	22.6892	4.9498
0.0806	22.1574	38000	0.1372	22.5506	4.8653
0.084	22.7405	39000	0.1350	22.3530	4.8049
0.0825	23.3236	40000	0.1380	22.1812	4.7661
0.0788	23.9067	41000	0.1407	22.1221	4.7596
0.0632	24.4898	42000	0.1385	22.0463	4.7328
0.0502	25.0729	43000	0.1393	21.9761	4.6974
0.0482	25.6560	44000	0.1384	21.9946	4.6878
0.0612	26.2391	45000	0.1364	21.7619	4.6330
0.0708	26.8222	46000	0.1346	21.7378	4.6234
0.0609	27.4052	47000	0.1353	21.7101	4.6142
0.0632	27.9883	48000	0.1361	21.7064	4.6071
0.0528	28.5714	49000	0.1350	21.6769	4.6064
0.0696	29.1545	50000	0.1347	21.6381	4.5898
0.0564	29.7376	51000	0.1348	21.6270	4.5904

Framework versions

• Transformers 4.48.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0