wav2vec2-xlsr-1b-mecita-portuguese-all-grade-2-4

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: 0.1991
• Wer: 0.1167
• Cer: 0.0331

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
31.355	0.99	47	3.2072	1.0	1.0
31.355	2.0	95	2.9060	0.9876	0.9942
6.8531	2.99	142	1.2010	0.9949	0.4072
6.8531	4.0	190	0.3287	0.2046	0.0592
1.5035	4.99	237	0.2646	0.1556	0.0453
1.5035	6.0	285	0.2319	0.1347	0.0392
0.4151	6.99	332	0.2259	0.1280	0.0366
0.4151	8.0	380	0.2191	0.1297	0.0352
0.3173	8.99	427	0.2036	0.1206	0.0346
0.3173	10.0	475	0.2189	0.1246	0.0353
0.2376	10.99	522	0.2133	0.1206	0.0333
0.2376	12.0	570	0.2189	0.1167	0.0326
0.2298	12.99	617	0.1991	0.1167	0.0331
0.2298	14.0	665	0.2027	0.1105	0.0307
0.1984	14.99	712	0.2037	0.1150	0.0315
0.1984	16.0	760	0.2268	0.1094	0.0328
0.1739	16.99	807	0.2252	0.1218	0.0341
0.1739	18.0	855	0.2075	0.1161	0.0330
0.156	18.99	902	0.2142	0.1088	0.0316
0.156	20.0	950	0.2155	0.1065	0.0328
0.156	20.99	997	0.2072	0.1099	0.0307
0.1493	22.0	1045	0.2052	0.1116	0.0316
0.1493	22.99	1092	0.2074	0.1094	0.0298
0.1526	24.0	1140	0.2162	0.1094	0.0308
0.1526	24.99	1187	0.2260	0.1133	0.0323
0.1401	26.0	1235	0.2228	0.1139	0.0321
0.1401	26.99	1282	0.2394	0.1082	0.0325
0.1323	28.0	1330	0.2096	0.1060	0.0318
0.1323	28.99	1377	0.2272	0.1139	0.0330
0.135	30.0	1425	0.2158	0.1099	0.0330
0.135	30.99	1472	0.2170	0.1139	0.0337
0.1263	32.0	1520	0.2097	0.1094	0.0315
0.1263	32.99	1567	0.2043	0.1122	0.0326

Framework versions

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