xuliu15/FT-frisian-10h

Whisper Small Frisian 10h

This model is a fine-tuned version of openai/whisper-small on the Common Voice 6.1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3402
• Wer: 22.4274

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: 8
• 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: 50
• training_steps: 1500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0548	0.1070	100	1.0200	52.0620
0.6944	0.2139	200	0.7126	39.6222
0.6024	0.3209	300	0.6052	36.0791
0.4697	0.4278	400	0.5303	32.5040
0.4222	0.5348	500	0.4780	30.0766
0.4075	0.6417	600	0.4458	28.4691
0.374	0.7487	700	0.4151	26.9292
0.3381	0.8556	800	0.3949	25.4678
0.3235	0.9626	900	0.3764	24.8904
0.1861	1.0695	1000	0.3643	23.5716
0.1554	1.1765	1100	0.3608	23.4183
0.1639	1.2834	1200	0.3511	23.0298
0.1453	1.3904	1300	0.3449	22.6591
0.1531	1.4973	1400	0.3419	22.4452
0.1299	1.6043	1500	0.3402	22.4274

Framework versions

• Transformers 4.40.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1