whisper-small-tigrinya-r-28

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

• Cer: 16.3883
• Loss: 0.4323
• Wer: 44.4351

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
1.7237	0.6680	250	21.0719	0.4191	54.7073
1.2838	1.3340	500	16.5156	0.3613	46.5236
1.2354	2.0	750	15.6233	0.3347	44.5935
1.0210	2.6680	1000	15.6542	0.3255	44.0587
0.7947	3.3340	1250	15.2447	0.3284	43.1325
0.8205	4.0	1500	15.1736	0.3213	42.9502
0.6328	4.6680	1750	15.3297	0.3362	43.1205
0.4849	5.3340	2000	15.3243	0.3516	42.9173
0.5075	6.0	2250	15.2106	0.3511	42.5409
0.3795	6.6680	2500	15.8178	0.3769	43.8704
0.2632	7.3340	2750	16.1000	0.4020	43.9690
0.2492	8.0	3000	15.9667	0.4070	44.1125
0.1995	8.6680	3250	16.2856	0.4243	44.4471
0.1714	9.3340	3500	16.4018	0.4308	44.5397
0.1666	10.0	3750	16.3883	0.4323	44.4351

Framework versions

• PEFT 0.18.1
• Transformers 5.2.0
• Pytorch 2.9.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2