whisper-medium-fine-tune

This model is a fine-tuned version of aisha-org/whisper-large-v3-373k on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1236
• Wer Ortho: 10.9083
• Wer: 8.7833

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: 16
• seed: 42
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.3984	0.0577	500	0.2443	18.2211	14.1269
0.3147	0.1155	1000	0.2092	16.1001	12.7389
0.2601	0.1732	1500	0.1881	14.8083	11.7594
0.2569	0.2309	2000	0.1715	13.6158	10.8886
0.2136	0.2887	2500	0.1620	13.1851	10.4975
0.2055	0.3464	3000	0.1544	12.5482	9.9526
0.1756	0.4042	3500	0.1439	11.9068	9.3922
0.1909	0.4619	4000	0.1354	11.8848	9.6647
0.1813	0.5196	4500	0.1304	10.8230	8.5942
0.1513	0.5774	5000	0.1236	10.9083	8.7833

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1