wav2vec2-large-xlsr-mecita-coraa-portuguese-aug-random-all-03

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: 0.1357
• Wer: 0.0844
• Cer: 0.0268

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
2.883	1.0	514	2.5884	0.9955	0.9943
0.9481	2.0	1029	0.2560	0.1474	0.0436
0.742	3.0	1543	0.1802	0.1098	0.0340
0.625	4.0	2058	0.1590	0.0975	0.0308
0.6001	5.0	2572	0.1486	0.0887	0.0292
0.5208	6.0	3087	0.1424	0.0918	0.0284
0.4857	7.0	3601	0.1357	0.0844	0.0268
0.4458	8.0	4116	0.1375	0.0882	0.0317
0.4158	9.0	4630	0.1411	0.0839	0.0303
0.3915	10.0	5145	0.1457	0.0915	0.0319
0.3898	11.0	5659	0.1464	0.0870	0.0310
0.3562	12.0	6174	0.1500	0.0875	0.0314
0.3619	13.0	6688	0.1523	0.0877	0.0313
0.3283	14.0	7203	0.1473	0.0856	0.0290
0.3196	15.0	7717	0.1443	0.0844	0.0299
0.3165	16.0	8232	0.1413	0.0813	0.0283
0.2954	17.0	8746	0.1451	0.0825	0.0283
0.293	18.0	9261	0.1539	0.0822	0.0286
0.2821	19.0	9775	0.1552	0.0844	0.0296
0.2893	20.0	10290	0.1484	0.0820	0.0285
0.2609	21.0	10804	0.1636	0.0851	0.0307
0.2526	22.0	11319	0.1520	0.0856	0.0292
0.2571	23.0	11833	0.1449	0.0851	0.0291
0.2486	24.0	12348	0.1574	0.0865	0.0307
0.2501	25.0	12862	0.1490	0.0856	0.0295
0.2525	26.0	13377	0.1508	0.0827	0.0294
0.2452	27.0	13891	0.1511	0.0808	0.0290

Framework versions

• Transformers 4.28.0
• Pytorch 2.5.0+cu121
• Datasets 3.1.0
• Tokenizers 0.13.3