whisper-small-sukuma

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

• Loss: 0.1995
• Wer: 13.27
• Cer: 3.53

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
• 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: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.2402	1.2920	500	0.2913	29.22	7.38
0.1006	2.5840	1000	0.1916	17.51	4.71
0.0549	3.8760	1500	0.1726	14.94	4.06
0.0110	5.1680	2000	0.1825	13.78	3.68
0.0050	6.4599	2500	0.1896	13.66	3.61
0.0019	7.7519	3000	0.1931	13.49	3.55
0.0019	9.0439	3500	0.1964	13.19	3.51
0.0011	10.3359	4000	0.1995	13.27	3.53

Framework versions

• Transformers 5.0.0
• Pytorch 2.10.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2