wav2vec2-xlsr-1b-mecita-portuguese-all-grade-4

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

• Loss: 0.1670
• Wer: 0.1139
• Cer: 0.0299

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: 3e-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: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
13.2754	0.93	7	3.4367	1.0	1.0
13.2754	2.0	15	2.9640	1.0	1.0
13.2754	2.93	22	2.8670	1.0	1.0
13.2754	4.0	30	2.8085	1.0	1.0
13.2754	4.93	37	2.7473	1.0	1.0
13.2754	6.0	45	2.4822	0.9975	0.9959
13.2754	6.93	52	2.0685	0.9975	0.7143
13.2754	8.0	60	1.1670	0.9975	0.4118
13.2754	8.93	67	0.5773	0.5025	0.1085
13.2754	10.0	75	0.3583	0.3094	0.0684
13.2754	10.93	82	0.2851	0.2030	0.0483
13.2754	12.0	90	0.2303	0.1832	0.0426
13.2754	12.93	97	0.2180	0.1485	0.0381
2.2909	14.0	105	0.2001	0.1386	0.0360
2.2909	14.93	112	0.1923	0.1262	0.0327
2.2909	16.0	120	0.1880	0.1213	0.0336
2.2909	16.93	127	0.1753	0.1238	0.0323
2.2909	18.0	135	0.1824	0.1139	0.0307
2.2909	18.93	142	0.1670	0.1139	0.0299
2.2909	20.0	150	0.1757	0.1064	0.0295
2.2909	20.93	157	0.1833	0.1114	0.0303
2.2909	22.0	165	0.1862	0.1238	0.0327
2.2909	22.93	172	0.1779	0.1163	0.0303
2.2909	24.0	180	0.1891	0.1114	0.0315
2.2909	24.93	187	0.2025	0.1188	0.0323
2.2909	26.0	195	0.2075	0.1238	0.0344
0.194	26.93	202	0.2085	0.1213	0.0340
0.194	28.0	210	0.1905	0.1163	0.0323
0.194	28.93	217	0.1793	0.1163	0.0327
0.194	30.0	225	0.1771	0.1114	0.0307
0.194	30.93	232	0.1784	0.1089	0.0295
0.194	32.0	240	0.1823	0.1188	0.0332
0.194	32.93	247	0.1797	0.1163	0.0319
0.194	34.0	255	0.1769	0.1114	0.0295
0.194	34.93	262	0.1740	0.1139	0.0307
0.194	36.0	270	0.1727	0.1139	0.0299
0.194	36.93	277	0.1738	0.1139	0.0287
0.194	38.0	285	0.1677	0.1064	0.0282
0.194	38.93	292	0.1694	0.1139	0.0295

Framework versions

• Transformers 4.28.0
• Pytorch 2.2.1+cu121
• Datasets 2.17.0
• Tokenizers 0.13.3