wav2vec2-large-xlsr-coraa-words-exp-5

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: 0.5522
• Wer: 0.3858
• Cer: 0.1683
• Per: 0.3792

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: 5e-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
37.2852	0.98	21	8.6659	1.0	0.9606	1.0
37.2852	2.0	43	4.1178	1.0	0.9606	1.0
37.2852	2.98	64	3.6499	1.0	0.9606	1.0
37.2852	4.0	86	3.4506	1.0	0.9606	1.0
9.8089	4.98	107	3.2834	1.0	0.9606	1.0
9.8089	6.0	129	3.1707	1.0	0.9606	1.0
9.8089	6.98	150	3.1196	1.0	0.9606	1.0
9.8089	8.0	172	3.0800	1.0	0.9606	1.0
9.8089	8.98	193	3.0656	1.0	0.9606	1.0
3.0165	10.0	215	3.0600	1.0	0.9606	1.0
3.0165	10.98	236	3.0494	1.0	0.9606	1.0
3.0165	12.0	258	3.0276	1.0	0.9606	1.0
3.0165	12.98	279	3.0055	1.0	0.9606	1.0
2.9518	14.0	301	2.8756	1.0	0.9606	1.0
2.9518	14.98	322	2.7060	1.0	0.9559	1.0
2.9518	16.0	344	2.3252	1.0	0.7022	1.0
2.9518	16.98	365	1.7212	1.0	0.5119	1.0
2.9518	18.0	387	1.1926	0.9981	0.3489	0.9981
2.0753	18.98	408	0.8904	0.6717	0.2300	0.6481
2.0753	20.0	430	0.7190	0.5038	0.1892	0.4896
2.0753	20.98	451	0.6968	0.4358	0.1814	0.4283
2.0753	22.0	473	0.6706	0.4	0.1756	0.3915
2.0753	22.98	494	0.6523	0.4104	0.1759	0.4038
0.6009	24.0	516	0.6112	0.4009	0.1725	0.3934
0.6009	24.98	537	0.6001	0.4019	0.1714	0.3943
0.6009	26.0	559	0.6171	0.3849	0.1713	0.3811
0.6009	26.98	580	0.5733	0.3877	0.1713	0.3821
0.3421	28.0	602	0.5823	0.3934	0.1710	0.3877
0.3421	28.98	623	0.5912	0.3877	0.1699	0.3802
0.3421	30.0	645	0.5539	0.3802	0.1664	0.3726
0.3421	30.98	666	0.5964	0.3745	0.1685	0.3679
0.3421	32.0	688	0.6107	0.3811	0.1692	0.3745
0.2715	32.98	709	0.5522	0.3858	0.1683	0.3792
0.2715	34.0	731	0.5745	0.3830	0.1693	0.3736
0.2715	34.98	752	0.6193	0.3830	0.1725	0.3764
0.2715	36.0	774	0.5876	0.3811	0.1690	0.3736
0.2715	36.98	795	0.5938	0.3858	0.1707	0.3802
0.2325	38.0	817	0.5815	0.3868	0.1710	0.3802
0.2325	38.98	838	0.5850	0.3802	0.1685	0.3736
0.2325	40.0	860	0.5995	0.3783	0.1697	0.3726
0.2325	40.98	881	0.5876	0.3783	0.1669	0.3708
0.2133	42.0	903	0.6039	0.3858	0.1693	0.3792
0.2133	42.98	924	0.5655	0.3755	0.1641	0.3651
0.2133	44.0	946	0.5673	0.3736	0.1655	0.3651
0.2133	44.98	967	0.6071	0.3613	0.1629	0.3528
0.2133	46.0	989	0.5673	0.3726	0.1650	0.3670
0.1754	46.98	1010	0.5796	0.3745	0.1636	0.3679
0.1754	48.0	1032	0.5837	0.3670	0.1652	0.3613
0.1754	48.98	1053	0.6056	0.3745	0.1678	0.3679
0.1754	50.0	1075	0.5833	0.3708	0.1644	0.3651
0.1754	50.98	1096	0.5853	0.3736	0.1655	0.3660
0.1569	52.0	1118	0.5841	0.3679	0.1631	0.3613
0.1569	52.98	1139	0.5640	0.3698	0.1620	0.3632

Framework versions

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