w2v-bert-2.0-igbo_naijavoices_1h

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the NAIJAVOICES_IGBO_1H - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.8551
• Wer: 0.6458
• Cer: 0.2352

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: 160
• eval_batch_size: 160
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 320
• total_eval_batch_size: 320
• optimizer: Use OptimizerNames.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_ratio: 0.1
• num_epochs: 25000.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
6.2706	33.3333	100	6.3346	1.8011	1.0293
3.6487	66.6667	200	3.6301	0.9877	0.8946
3.0158	100.0	300	3.0263	1.0	1.0000
2.3833	133.3333	400	2.2797	1.0013	0.6634
1.256	166.6667	500	1.1780	0.8874	0.3541
0.8752	200.0	600	0.9456	0.7528	0.2749
0.6649	233.3333	700	0.8731	0.6848	0.2482
0.52	266.6667	800	0.8531	0.6449	0.2352
0.411	300.0	900	0.8922	0.6298	0.2294
0.26	333.3333	1000	0.9620	0.6131	0.2253
0.1774	366.6667	1100	1.0638	0.6135	0.2232
0.1402	400.0	1200	1.1608	0.6048	0.2210
0.0775	433.3333	1300	1.2847	0.6022	0.2184

Framework versions

• Transformers 4.48.1
• Pytorch 2.7.1+cu126
• Datasets 4.0.0
• Tokenizers 0.21.2