wav2vec2-large-xlsr-coraa-words-exp-4

This model is a fine-tuned version of Edresson/wav2vec2-large-xlsr-coraa-portuguese on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5520
• Wer: 0.3792
• Cer: 0.1634
• Per: 0.3736

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: 4e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine_with_restarts
• num_epochs: 200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Per
36.9896	0.98	21	11.5637	1.0	0.9606	1.0
36.9896	2.0	43	4.4906	1.0	0.9606	1.0
36.9896	2.98	64	3.8521	1.0	0.9606	1.0
36.9896	4.0	86	3.6001	1.0	0.9606	1.0
10.4512	4.98	107	3.3694	1.0	0.9606	1.0
10.4512	6.0	129	3.2463	1.0	0.9606	1.0
10.4512	6.98	150	3.1842	1.0	0.9606	1.0
10.4512	8.0	172	3.1160	1.0	0.9606	1.0
10.4512	8.98	193	3.0945	1.0	0.9606	1.0
3.0577	10.0	215	3.0865	1.0	0.9606	1.0
3.0577	10.98	236	3.0657	1.0	0.9606	1.0
3.0577	12.0	258	3.0591	1.0	0.9606	1.0
3.0577	12.98	279	3.0367	1.0	0.9606	1.0
2.9732	14.0	301	3.0339	1.0	0.9606	1.0
2.9732	14.98	322	3.0231	1.0	0.9606	1.0
2.9732	16.0	344	2.9633	1.0	0.9606	1.0
2.9732	16.98	365	2.8554	1.0	0.9606	1.0
2.9732	18.0	387	2.7109	1.0	0.9606	1.0
2.8449	18.98	408	2.3843	1.0	0.7400	1.0
2.8449	20.0	430	1.8336	1.0	0.5582	1.0
2.8449	20.98	451	1.3493	1.0	0.3643	1.0
2.8449	22.0	473	1.0668	0.9991	0.3247	0.9991
2.8449	22.98	494	0.8907	0.9991	0.3030	0.9991
1.5487	24.0	516	0.8101	0.9972	0.2923	0.9972
1.5487	24.98	537	0.7335	0.9802	0.2799	0.9792
1.5487	26.0	559	0.7279	0.9623	0.2740	0.9623
1.5487	26.98	580	0.6539	0.5264	0.1914	0.5236
0.5979	28.0	602	0.6419	0.4472	0.1821	0.4387
0.5979	28.98	623	0.6303	0.4179	0.1775	0.4085
0.5979	30.0	645	0.6278	0.4047	0.1739	0.3962
0.5979	30.98	666	0.6249	0.3896	0.1727	0.3830
0.5979	32.0	688	0.6010	0.3962	0.1700	0.3868
0.3815	32.98	709	0.5864	0.3943	0.1710	0.3868
0.3815	34.0	731	0.5773	0.3887	0.1703	0.3811
0.3815	34.98	752	0.5817	0.3792	0.1693	0.3708
0.3815	36.0	774	0.5911	0.3887	0.1723	0.3792
0.3815	36.98	795	0.5546	0.3887	0.1671	0.3792
0.3023	38.0	817	0.5548	0.3868	0.1672	0.3774
0.3023	38.98	838	0.5555	0.3821	0.1673	0.3745
0.3023	40.0	860	0.5582	0.3783	0.1676	0.3689
0.3023	40.98	881	0.5644	0.3896	0.1682	0.3811
0.2719	42.0	903	0.5822	0.3830	0.1673	0.3774
0.2719	42.98	924	0.5649	0.3811	0.1652	0.3717
0.2719	44.0	946	0.5553	0.3774	0.1641	0.3698
0.2719	44.98	967	0.5777	0.3745	0.1654	0.3660
0.2719	46.0	989	0.5868	0.3792	0.1664	0.3726
0.2202	46.98	1010	0.5546	0.3670	0.1626	0.3594
0.2202	48.0	1032	0.5724	0.3632	0.1624	0.3557
0.2202	48.98	1053	0.5851	0.3698	0.1657	0.3613
0.2202	50.0	1075	0.5671	0.3774	0.1637	0.3689
0.2202	50.98	1096	0.5720	0.3726	0.1651	0.3642
0.1995	52.0	1118	0.5672	0.3755	0.1644	0.3670
0.1995	52.98	1139	0.5520	0.3792	0.1634	0.3736
0.1995	54.0	1161	0.5687	0.3613	0.1617	0.3557
0.1995	54.98	1182	0.5597	0.3623	0.1598	0.3566
0.1802	56.0	1204	0.5828	0.3594	0.1605	0.3528
0.1802	56.98	1225	0.5753	0.3717	0.1636	0.3642
0.1802	58.0	1247	0.5592	0.3613	0.1627	0.3547
0.1802	58.98	1268	0.5694	0.3679	0.1650	0.3632
0.1802	60.0	1290	0.6006	0.3679	0.1640	0.3613
0.1642	60.98	1311	0.6042	0.3698	0.1644	0.3613
0.1642	62.0	1333	0.5754	0.3651	0.1633	0.3566
0.1642	62.98	1354	0.5793	0.3670	0.1630	0.3566
0.1642	64.0	1376	0.5941	0.3660	0.1627	0.3557
0.1642	64.98	1397	0.5825	0.3774	0.1624	0.3679
0.1466	66.0	1419	0.5996	0.3689	0.1627	0.3613
0.1466	66.98	1440	0.5777	0.3745	0.1631	0.3679
0.1466	68.0	1462	0.5910	0.3679	0.1620	0.3604
0.1466	68.98	1483	0.5821	0.3698	0.1626	0.3623
0.1398	70.0	1505	0.5960	0.3679	0.1643	0.3613
0.1398	70.98	1526	0.5922	0.3604	0.1640	0.3519
0.1398	72.0	1548	0.5987	0.3698	0.1633	0.3623
0.1398	72.98	1569	0.6110	0.3736	0.1654	0.3651

Framework versions

• Transformers 4.28.0
• Pytorch 2.4.1+cu121
• Datasets 3.2.0
• Tokenizers 0.13.3