zirri23/whisper-akan-finetuned

Whisper Akan Finetuned

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

• Loss: 0.5045
• Wer: 38.8906

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.5849	0.8795	500	0.7880	58.7006
1.1377	1.7581	1000	0.5842	45.0593
0.9672	2.6368	1500	0.5044	38.2501
0.8439	3.5154	2000	0.4759	37.6913
0.7449	4.3940	2500	0.4575	42.1849
0.5790	5.2726	3000	0.4555	40.9956
0.5198	6.1513	3500	0.4545	46.8275
0.5027	7.0299	4000	0.4539	47.9825

Framework versions

• PEFT 0.18.1
• Transformers 5.0.0
• Pytorch 2.9.0+cu126
• Datasets 4.5.0
• Tokenizers 0.22.2