whisper-small-lingala-cased-2

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

• Loss: 0.0792
• Wer Keep: 0.0783
• Cer Keep: 0.0268
• Zindi Keep: 0.9475
• Zindi Strip: 0.9639
• Zindi Lower: 0.9717
• Pct Capitalised: 0.7617
• Pct Punctuated: 0.74

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: 200
• num_epochs: 5.0
• 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
0.1264	0.4357	200	0.0646	0.0681	0.0272	0.9523	0.9651	0.9709	0.7583	0.7433
0.1359	0.8715	400	0.0690	0.0724	0.0267	0.9504	0.9638	0.9707	0.7433	0.7067
0.0656	1.3072	600	0.0775	0.0793	0.0296	0.9456	0.9614	0.9681	0.7883	0.82
0.0670	1.7429	800	0.0765	0.0810	0.0281	0.9454	0.9631	0.9707	0.825	0.805
0.0248	2.1786	1000	0.0791	0.0764	0.0259	0.9489	0.9637	0.9705	0.7933	0.7967
0.0187	2.6144	1200	0.0792	0.0783	0.0268	0.9475	0.9639	0.9717	0.7617	0.74

Framework versions

• Transformers 5.14.1
• Pytorch 2.13.0+cu130
• Datasets 3.6.0
• Tokenizers 0.22.2