whisper-small-tigrinya-r-32

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: 16.2664
• Loss: 0.4443
• Wer: 44.0975

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
1.7067	0.6680	250	20.2379	0.4194	53.5989
1.2662	1.3340	500	16.9180	0.3535	47.0435
1.2181	2.0	750	15.7577	0.3320	44.4949
1.0008	2.6680	1000	15.0975	0.3233	42.9323
0.7737	3.3340	1250	15.4744	0.3279	43.3595
0.8006	4.0	1500	14.7349	0.3221	42.7859
0.6077	4.6680	1750	15.3030	0.3387	42.8725
0.4587	5.3340	2000	15.5312	0.3529	43.2012
0.4760	6.0	2250	15.4100	0.3548	42.8665
0.3458	6.6680	2500	16.5764	0.3845	44.9252
0.2320	7.3340	2750	16.5760	0.4112	44.6443
0.2221	8.0	3000	16.3563	0.4165	44.2589
0.1678	8.6680	3250	16.8778	0.4360	45.0865
0.1415	9.3340	3500	16.4335	0.4432	44.2230
0.1373	10.0	3750	16.2664	0.4443	44.0975

Framework versions

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