wav2vec2-large-xlsr-grosman-1b-words-aug-exp-1

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: 17.0754
• Wer: 1.0
• Cer: 0.9215
• Per: 1.0

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: 200
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Per
18.8645	1.0	126	17.0754	1.0	0.9215	1.0
6.2004	2.0	253	27.5514	1.0019	0.9311	1.0019
4.7666	3.0	379	26.1635	1.0019	0.9187	1.0019
4.1703	4.0	506	29.2433	1.0009	0.8744	1.0009
3.8788	5.0	632	26.7764	1.0123	0.7640	1.0123
3.4295	6.0	759	24.4870	1.0057	0.7191	1.0057
3.4233	7.0	885	27.2557	1.0283	0.6642	1.0283
2.85	8.0	1012	20.5990	0.9953	0.5631	0.9877
2.4174	9.0	1138	24.5312	0.8594	0.4160	0.8566
1.8357	10.0	1265	24.3372	0.6840	0.3444	0.6821
1.5625	11.0	1391	25.7571	0.6123	0.3061	0.6057
1.5615	12.0	1518	23.9230	0.5811	0.2910	0.5726
1.087	13.0	1644	26.9014	0.5660	0.2785	0.5566
0.9967	14.0	1771	27.5072	0.5868	0.2811	0.5783
1.0407	15.0	1897	24.3373	0.5481	0.2662	0.5396
0.8217	16.0	2024	27.1505	0.4925	0.2625	0.4858
0.8024	17.0	2150	26.4263	0.5208	0.2644	0.5123
0.763	18.0	2277	25.0467	0.4792	0.2562	0.4717
0.6881	19.0	2403	26.9578	0.5019	0.2639	0.4934
0.5914	20.0	2530	26.7000	0.4679	0.2552	0.4594
0.6033	21.0	2656	28.2333	0.4623	0.2453	0.4547

Framework versions

• Transformers 4.28.0
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.13.3