gikuyu-asr-v0

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

• Loss: 0.2944
• Wer: 32.8628
• Cer: 10.4740

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: 32
• eval_batch_size: 16
• 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
• num_epochs: 2
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.5621	0.2116	500	0.4916	53.2794	16.7317
0.4414	0.4232	1000	0.3852	42.3825	13.5806
0.3914	0.6348	1500	0.3505	38.0444	12.2589
0.3829	0.8464	2000	0.3302	36.8910	12.0260
0.3188	1.0580	2500	0.3157	34.6703	11.7171
0.3146	1.2696	3000	0.3089	34.4293	11.1204
0.2946	1.4812	3500	0.3040	33.9301	10.8692
0.3036	1.6928	4000	0.2974	33.5170	10.8927
0.2972	1.9044	4500	0.2952	32.8972	10.4897
0.2979	2.0	4726	0.2944	32.8628	10.4740

Framework versions

• Transformers 5.15.0
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.22.2