MODELING_23

This model is a fine-tuned version of openai/whisper-small on the abdouaziiz/full_wolof_normalized dataset. It achieves the following results on the evaluation set:

• Loss: 0.4923
• Wer: 0.2679

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: 16
• seed: 42
• gradient_accumulation_steps: 2
• 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: 50
• training_steps: 48000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.9983	0.2723	500	0.7120	0.6396
1.2974	0.5445	1000	0.6105	0.5783
1.1407	0.8168	1500	0.5547	0.5724
0.9514	1.0890	2000	0.5317	0.4652
0.7178	1.3613	2500	0.5096	0.4635
0.7264	1.6335	3000	0.4975	0.3532
0.7091	1.9058	3500	0.4781	0.3160
0.5207	2.1781	4000	0.4779	0.3010
0.4213	2.4503	4500	0.4792	0.4107
0.442	2.7226	5000	0.4751	0.3008
0.4337	2.9948	5500	0.4598	0.3062
0.2341	3.2671	6000	0.4918	0.3113
0.2406	3.5393	6500	0.4923	0.2679
0.2479	3.8116	7000	0.4866	0.2887
0.2078	4.0839	7500	0.5019	0.2913
0.1258	4.3561	8000	0.5140	0.2730

Framework versions

• Transformers 4.46.0
• Pytorch 2.7.0+cu126
• Datasets 3.3.2
• Tokenizers 0.20.3