wav2vec2-bert-swahili-noise

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

• Loss: 0.5771
• Wer: 0.2959

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: 1e-05
• train_batch_size: 38
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_BNB 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: 100
• training_steps: 900

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.142	0.0817	100	3.0798	0.9999
1.9368	0.1634	200	1.4137	0.9886
0.9455	0.2451	300	0.7745	0.4115
0.8236	0.3268	400	0.6644	0.3485
0.7723	0.4085	500	0.6475	0.3313
0.7603	0.4902	600	0.6082	0.3097
0.6848	0.5719	700	0.5972	0.3072
0.683	0.6536	800	0.5762	0.2986
0.6967	0.7353	900	0.5771	0.2959

Framework versions

• Transformers 4.50.3
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.21.2