lyimo/whisper-small-sw-badili-v4

Whisper Small Swahili - Badili

This model is a fine-tuned version of openai/whisper-small on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4912
• Wer: 739.2268

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 12000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4265	0.43	1000	0.5471	43.0938
0.34	0.87	2000	0.4611	46.8067
0.2026	1.3	3000	0.4447	226.8087
0.1963	1.74	4000	0.4141	134.1173
0.1258	2.17	5000	0.4238	380.2596
0.1176	2.61	6000	0.4271	355.4544
0.0646	3.04	7000	0.4334	443.7180
0.0619	3.47	8000	0.4460	432.6712
0.0648	3.91	9000	0.4421	366.3092
0.0359	4.34	10000	0.4681	511.0789
0.029	4.78	11000	0.4795	668.5806
0.0179	5.21	12000	0.4912	739.2268

Framework versions

• Transformers 4.27.0.dev0
• Pytorch 1.13.1+cu117
• Datasets 2.10.1
• Tokenizers 0.13.2