wav2vec2-large-xlsr-coraa-words-exp-3

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: 0.5878
• Wer: 0.3943
• Cer: 0.1664
• Per: 0.3858

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: linear
• num_epochs: 200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Per
37.3293	0.98	21	11.1197	1.0	0.9606	1.0
37.3293	2.0	43	4.4296	1.0	0.9606	1.0
37.3293	2.98	64	3.8433	1.0	0.9606	1.0
37.3293	4.0	86	3.6004	1.0	0.9606	1.0
10.4132	4.98	107	3.4505	1.0	0.9606	1.0
10.4132	6.0	129	3.2447	1.0	0.9606	1.0
10.4132	6.98	150	3.1682	1.0	0.9606	1.0
10.4132	8.0	172	3.1254	1.0	0.9606	1.0
10.4132	8.98	193	3.1141	1.0	0.9606	1.0
3.0609	10.0	215	3.0941	1.0	0.9606	1.0
3.0609	10.98	236	3.0881	1.0	0.9606	1.0
3.0609	12.0	258	3.0595	1.0	0.9606	1.0
3.0609	12.98	279	3.0402	1.0	0.9606	1.0
2.972	14.0	301	3.0408	1.0	0.9606	1.0
2.972	14.98	322	3.0179	1.0	0.9606	1.0
2.972	16.0	344	2.9057	1.0	0.9606	1.0
2.972	16.98	365	2.7601	1.0	0.9606	1.0
2.972	18.0	387	2.6964	1.0	0.9384	1.0
2.7812	18.98	408	2.3877	1.0	0.7817	1.0
2.7812	20.0	430	1.8978	1.0	0.6250	1.0
2.7812	20.98	451	1.4096	1.0	0.4215	1.0
2.7812	22.0	473	1.0923	0.9991	0.3336	0.9991
2.7812	22.98	494	0.9002	0.9991	0.3079	0.9991
1.4929	24.0	516	0.8111	0.9991	0.2955	0.9991
1.4929	24.98	537	0.7687	0.9953	0.2888	0.9953
1.4929	26.0	559	0.7479	0.9811	0.2749	0.9811
1.4929	26.98	580	0.7179	0.9698	0.2690	0.9689
0.5434	28.0	602	0.6987	0.8028	0.2307	0.8009
0.5434	28.98	623	0.6984	0.4547	0.1811	0.4453
0.5434	30.0	645	0.6479	0.4283	0.1760	0.4208
0.5434	30.98	666	0.6504	0.4387	0.1782	0.4302
0.5434	32.0	688	0.6665	0.4217	0.1741	0.4094
0.3255	32.98	709	0.6688	0.4292	0.1763	0.4208
0.3255	34.0	731	0.6182	0.4179	0.1725	0.4066
0.3255	34.98	752	0.6193	0.4066	0.1718	0.4
0.3255	36.0	774	0.6320	0.4085	0.1716	0.3981
0.3255	36.98	795	0.6316	0.4075	0.1718	0.4009
0.231	38.0	817	0.6636	0.3991	0.1713	0.3896
0.231	38.98	838	0.6038	0.3953	0.1679	0.3887
0.231	40.0	860	0.6158	0.4	0.1680	0.3915
0.231	40.98	881	0.6122	0.3896	0.1672	0.3821
0.2015	42.0	903	0.6088	0.3877	0.1682	0.3830
0.2015	42.98	924	0.5878	0.3943	0.1664	0.3858
0.2015	44.0	946	0.6330	0.3811	0.1662	0.3764
0.2015	44.98	967	0.6317	0.3887	0.1664	0.3821
0.2015	46.0	989	0.6155	0.3906	0.1671	0.3858
0.1617	46.98	1010	0.6348	0.3811	0.1648	0.3774
0.1617	48.0	1032	0.6249	0.3840	0.1655	0.3774
0.1617	48.98	1053	0.6249	0.3915	0.1669	0.3868
0.1617	50.0	1075	0.6070	0.3877	0.1659	0.3811
0.1617	50.98	1096	0.6005	0.3849	0.1637	0.3764
0.1426	52.0	1118	0.6131	0.3868	0.1638	0.3802
0.1426	52.98	1139	0.6216	0.3925	0.1668	0.3868
0.1426	54.0	1161	0.6343	0.3953	0.1672	0.3849
0.1426	54.98	1182	0.6267	0.3896	0.1643	0.3821
0.1273	56.0	1204	0.6093	0.3849	0.1647	0.3792
0.1273	56.98	1225	0.6199	0.3906	0.1654	0.3821
0.1273	58.0	1247	0.5931	0.3915	0.1659	0.3830
0.1273	58.98	1268	0.6227	0.3877	0.1662	0.3802
0.1273	60.0	1290	0.6208	0.3783	0.1652	0.3698
0.1106	60.98	1311	0.6312	0.3802	0.1647	0.3745
0.1106	62.0	1333	0.6203	0.3830	0.1671	0.3764
0.1106	62.98	1354	0.6219	0.3915	0.1673	0.3830

Framework versions

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