wav2vec2-xls-r-1b-portuguese-casa-civil-030124

This model is a fine-tuned version of jonatasgrosman/wav2vec2-xls-r-1b-portuguese on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5479
• Wer: 0.1310

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: 0.0003
• 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
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
28.5913	2.0	100	1.2903	0.1460
1.1869	4.0	200	0.6083	0.1537
0.8173	6.0	300	0.7054	0.2217
0.7882	8.0	400	0.7377	0.2711
0.6783	10.0	500	0.7785	0.2321
0.5541	12.0	600	0.6881	0.2394
0.5104	14.0	700	0.7285	0.2270
0.344	16.0	800	0.6114	0.1991
0.304	18.0	900	0.5559	0.1906
0.2315	20.0	1000	0.6833	0.1727
0.2144	22.0	1100	0.5632	0.1695
0.1725	24.0	1200	0.5597	0.1463
0.1492	26.0	1300	0.5356	0.1472
0.118	28.0	1400	0.5499	0.1344
0.1083	30.0	1500	0.5479	0.1310

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2+cu121
• Datasets 2.16.0
• Tokenizers 0.15.0