wav2vec2-xlsr-1b-mecita-portuguese-all-grade-2-5

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: 0.2238
• Wer: 0.1009
• Cer: 0.0334

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
31.1421	1.0	58	7.4643	1.0	0.9272
8.4474	2.0	116	2.9703	0.9992	0.9997
8.4474	3.0	174	2.8949	1.0	1.0
2.9538	4.0	232	2.2467	0.9973	0.9806
2.9538	5.0	290	0.4163	0.2921	0.0764
1.7017	6.0	348	0.2723	0.1421	0.0436
0.4188	7.0	406	0.2676	0.1280	0.0421
0.4188	8.0	464	0.2381	0.1127	0.0379
0.3001	9.0	522	0.2585	0.1127	0.0361
0.3001	10.0	580	0.2487	0.1084	0.0359
0.2349	11.0	638	0.2271	0.1029	0.0332
0.2349	12.0	696	0.2238	0.1009	0.0334
0.2051	13.0	754	0.2451	0.1025	0.0333
0.1752	14.0	812	0.2595	0.1029	0.0343
0.1752	15.0	870	0.2571	0.1048	0.0350
0.1839	16.0	928	0.2540	0.1044	0.0344
0.1839	17.0	986	0.2417	0.1064	0.0357
0.1664	18.0	1044	0.2350	0.1005	0.0331
0.1796	19.0	1102	0.2415	0.0989	0.0341
0.1796	20.0	1160	0.2348	0.0989	0.0334
0.1446	21.0	1218	0.2538	0.0958	0.0327
0.1446	22.0	1276	0.2357	0.0954	0.0317
0.1518	23.0	1334	0.2329	0.0911	0.0314
0.1518	24.0	1392	0.2412	0.0989	0.0322
0.1185	25.0	1450	0.2365	0.0919	0.0310
0.1273	26.0	1508	0.2390	0.0958	0.0314
0.1273	27.0	1566	0.2380	0.0946	0.0318
0.115	28.0	1624	0.2333	0.0970	0.0322
0.115	29.0	1682	0.2393	0.0934	0.0313
0.1244	30.0	1740	0.2354	0.0974	0.0319
0.1244	31.0	1798	0.2622	0.0938	0.0323
0.1189	32.0	1856	0.2796	0.0982	0.0322

Framework versions

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