Ajegetina/afrispeech_ayarma_small

Whisper small - Ayarma

This model is a fine-tuned version of openai/whisper-small on the Afrispeech 200 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3855
• Wer: 14.7658

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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_steps: 100
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5725	0.1667	500	0.4557	16.7119
0.5284	0.3333	1000	0.4427	17.2211
0.5196	0.5	1500	0.4068	15.8407
0.3895	1.075	2000	0.3898	15.2071
0.3543	1.2417	2500	0.3841	15.1392
0.4652	1.4083	3000	0.3855	14.7658

Framework versions

• Transformers 4.52.0
• Pytorch 2.9.0+cu126
• Datasets 2.19.0
• Tokenizers 0.21.4