whisper

This model is a fine-tuned version of openai/whisper-tiny on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.8306
• Wer: 70.7248

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• 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: 6000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4489	3.0143	500	1.2780	58.5381
0.1305	7.0057	1000	1.3309	59.5209
0.0447	10.02	1500	1.4363	70.3563
0.0148	14.0113	2000	1.4946	58.0344
0.0027	18.0027	2500	1.7187	68.3415
0.0011	21.017	3000	1.6621	64.8280
0.0005	25.0083	3500	1.7632	72.1622
0.0003	28.0227	4000	1.7390	69.1892
0.0003	32.014	4500	1.7954	71.9042
0.0002	36.0053	5000	1.8321	71.6462
0.0002	39.0197	5500	1.8165	70.1106
0.0002	43.011	6000	1.8306	70.7248

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.1
• Tokenizers 0.21.0