aipanjab/whisper-tiny-pa

Whisper-tiny

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

• Loss: 0.2197
• Wer: 44.9001

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
0.2708	1.2804	1000	0.2786	60.7745
0.1963	2.5608	2000	0.2130	51.0429
0.1617	3.8412	3000	0.1934	47.3317
0.1125	5.1216	4000	0.1897	45.6335
0.0968	6.4020	5000	0.1915	45.0640
0.0857	7.6825	6000	0.1949	44.7259
0.0757	8.9629	7000	0.1985	44.4762
0.0568	10.2433	8000	0.2107	44.6765
0.0531	11.5237	9000	0.2165	44.8611
0.05	12.8041	10000	0.2197	44.9001

Framework versions

• Transformers 4.46.3
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.3