whisper-small-shona-cased

This model is a fine-tuned version of sulaimank/whisper-small-shona-100hrs on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0195
• Wer Keep: 0.0289
• Cer Keep: 0.0043
• Zindi Keep: 0.9834
• Zindi Strip: 0.9906
• Zindi Lower: 0.9963
• Pct Capitalised: 0.9967
• Pct Punctuated: 0.9967

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: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Keep	Cer Keep	Zindi Keep	Zindi Strip	Zindi Lower	Pct Capitalised	Pct Punctuated
0.1181	0.4444	500	0.0884	0.1215	0.0186	0.9299	0.9575	0.9793	0.9983	0.9933
0.0874	0.8889	1000	0.0605	0.0946	0.0141	0.9457	0.9687	0.9865	1.0	0.9883
0.0491	1.3333	1500	0.0463	0.0746	0.0107	0.9573	0.9757	0.9896	0.9983	0.9983
0.0393	1.7778	2000	0.0331	0.0583	0.0084	0.9666	0.9811	0.9925	0.9967	0.99
0.0174	2.2222	2500	0.0242	0.0398	0.0058	0.9772	0.9873	0.9954	0.9983	1.0
0.0143	2.6667	3000	0.0195	0.0289	0.0043	0.9834	0.9906	0.9963	0.9967	0.9967

Framework versions

• Transformers 4.57.6
• Pytorch 2.11.0+cu128
• Datasets 3.6.0
• Tokenizers 0.22.2