Wav2vec2-wolof

This model is a fine-tuned version of asr-africa/wav2vec2-xls-r-wolof-mixed-75-hours on the LAFRICAMOBILE/FULL_WOLOF2 - DEFAULT dataset. It achieves the following results on the evaluation set:

• Loss: 0.2882
• Wer: 0.4645
• Cer: 0.1419

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 60.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.6813	0.2681	500	0.4663	0.5100	0.1591
0.5015	0.5362	1000	0.3608	0.4816	0.1488
0.4424	0.8043	1500	0.3339	0.4802	0.1456
0.3254	1.0724	2000	0.3147	0.4792	0.1471
0.5671	1.3405	2500	0.3150	0.4757	0.1470
0.3204	1.6086	3000	0.3072	0.4692	0.1465
0.32	1.8767	3500	0.2975	0.4685	0.1412
0.3854	2.1448	4000	0.3000	0.4664	0.1416
0.3809	2.4129	4500	0.2893	0.4607	0.1398
0.3977	2.6810	5000	0.2882	0.4644	0.1419
0.3966	2.9491	5500	0.3023	0.4657	0.1417
0.3585	3.2172	6000	0.3507	0.4764	0.1458
0.338	3.4853	6500	0.2906	0.4646	0.1412

Framework versions

• Transformers 4.53.3
• Pytorch 2.7.1+cu126
• Datasets 3.6.0
• Tokenizers 0.21.2