wav2vec2-large-xlsr-coraa-words-exp-2

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: 0.5347
• Wer: 0.3764
• Cer: 0.1666
• Per: 0.3689

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
• num_epochs: 200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Per
37.5783	0.98	21	12.5136	1.0	0.9606	1.0
37.5783	2.0	43	4.6324	1.0	0.9606	1.0
37.5783	2.98	64	3.9199	1.0	0.9606	1.0
37.5783	4.0	86	3.6333	1.0	0.9606	1.0
10.8343	4.98	107	3.3986	1.0	0.9606	1.0
10.8343	6.0	129	3.2632	1.0	0.9606	1.0
10.8343	6.98	150	3.1944	1.0	0.9606	1.0
10.8343	8.0	172	3.1359	1.0	0.9606	1.0
10.8343	8.98	193	3.1062	1.0	0.9606	1.0
3.0686	10.0	215	3.0914	1.0	0.9606	1.0
3.0686	10.98	236	3.0769	1.0	0.9606	1.0
3.0686	12.0	258	3.0595	1.0	0.9606	1.0
3.0686	12.98	279	3.0485	1.0	0.9606	1.0
2.9771	14.0	301	3.0664	1.0	0.9606	1.0
2.9771	14.98	322	3.0457	1.0	0.9606	1.0
2.9771	16.0	344	3.0246	1.0	0.9606	1.0
2.9771	16.98	365	3.0020	1.0	0.9606	1.0
2.9771	18.0	387	2.9227	1.0	0.9606	1.0
2.9246	18.98	408	2.8106	1.0	0.9606	1.0
2.9246	20.0	430	2.6933	1.0	0.9172	1.0
2.9246	20.98	451	2.4785	1.0	0.8042	1.0
2.9246	22.0	473	2.1101	1.0	0.6433	1.0
2.9246	22.98	494	1.6137	1.0	0.5067	1.0
2.3792	24.0	516	1.1653	0.9991	0.3579	0.9991
2.3792	24.98	537	0.9528	0.9991	0.3145	0.9991
2.3792	26.0	559	0.8157	0.9991	0.3004	0.9991
2.3792	26.98	580	0.7607	0.9962	0.2891	0.9962
0.8858	28.0	602	0.7309	0.9783	0.2836	0.9783
0.8858	28.98	623	0.6736	0.6830	0.2156	0.6811
0.8858	30.0	645	0.6378	0.4274	0.1760	0.4198
0.8858	30.98	666	0.5986	0.4217	0.1756	0.4151
0.8858	32.0	688	0.6263	0.3934	0.1721	0.3840
0.4514	32.98	709	0.6369	0.4189	0.1777	0.4123
0.4514	34.0	731	0.5785	0.3887	0.1713	0.3802
0.4514	34.98	752	0.5779	0.3925	0.1713	0.3840
0.4514	36.0	774	0.5766	0.3792	0.1682	0.3717
0.4514	36.98	795	0.5598	0.3755	0.1654	0.3670
0.3178	38.0	817	0.5604	0.3698	0.1662	0.3632
0.3178	38.98	838	0.5347	0.3764	0.1666	0.3689
0.3178	40.0	860	0.5741	0.3726	0.1683	0.3642
0.3178	40.98	881	0.5516	0.3736	0.1685	0.3660
0.2722	42.0	903	0.5712	0.3679	0.1669	0.3623
0.2722	42.98	924	0.5592	0.3755	0.1666	0.3670
0.2722	44.0	946	0.5517	0.3679	0.1648	0.3613
0.2722	44.98	967	0.5622	0.3566	0.1627	0.35
0.2722	46.0	989	0.5663	0.3651	0.1659	0.3575
0.2191	46.98	1010	0.5639	0.3679	0.1652	0.3623
0.2191	48.0	1032	0.5749	0.3538	0.1640	0.3453
0.2191	48.98	1053	0.5537	0.3642	0.1633	0.3566
0.2191	50.0	1075	0.5536	0.3736	0.1662	0.3679
0.2191	50.98	1096	0.5671	0.3774	0.1669	0.3717
0.1935	52.0	1118	0.5730	0.3670	0.1661	0.3594
0.1935	52.98	1139	0.5711	0.3717	0.1659	0.3632
0.1935	54.0	1161	0.5680	0.3623	0.1634	0.3528
0.1935	54.98	1182	0.5692	0.3575	0.1630	0.35
0.1755	56.0	1204	0.6076	0.3651	0.1675	0.3585
0.1755	56.98	1225	0.5917	0.3651	0.1655	0.3575
0.1755	58.0	1247	0.5720	0.3566	0.1636	0.3519
0.1755	58.98	1268	0.5759	0.3651	0.1641	0.3594

Framework versions

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