Whisper-large-v3-turbo-fa - Sadegh Karimi

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0839
• Wer: 9.6275

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: Use adamw_torch 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: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1789	0.0217	500	0.2427	26.4099
0.2077	0.0435	1000	0.2296	27.1873
0.1928	0.0652	1500	0.2320	27.5951
0.1801	0.0869	2000	0.2026	24.0409
0.1865	0.1086	2500	0.1925	22.3742
0.1535	0.1304	3000	0.1872	22.9511
0.1463	0.1521	3500	0.1786	21.5436
0.0935	0.1738	4000	0.1749	20.5330
0.1052	0.1956	4500	0.1597	19.0314
0.091	0.2173	5000	0.1553	20.2125
0.0743	0.2390	5500	0.1474	16.9160
0.096	0.2607	6000	0.1352	15.9027
0.111	0.2825	6500	0.1259	14.9071
0.089	0.3042	7000	0.1179	14.1146
0.0813	0.3259	7500	0.1101	12.8653
0.072	0.3477	8000	0.1012	11.8138
0.0715	0.3694	8500	0.0948	10.9791
0.0683	0.3911	9000	0.0903	10.2563
0.0634	0.4128	9500	0.0861	9.6616
0.0739	0.4346	10000	0.0839	9.6275

Framework versions

• Transformers 4.47.1
• Pytorch 2.1.0+cu118
• Datasets 3.2.0
• Tokenizers 0.21.0