anveshg03/whisper-tiny-finetuned

Whisper tiny eng

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

• Loss: 0.3574
• Wer: 5.5556

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
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant
• lr_scheduler_warmup_steps: 5
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.1546	1.0	3	1.6919	37.5
1.3508	2.0	6	1.1947	16.6667
0.9872	3.0	9	0.9060	12.5
0.782	4.0	12	0.7305	9.7222
0.6452	5.0	15	0.6134	8.3333
0.5597	6.0	18	0.5442	5.5556
0.5109	7.0	21	0.5086	5.5556
0.4854	8.0	24	0.4854	5.5556
0.4685	9.0	27	0.4690	5.5556
0.4558	10.0	30	0.4558	5.5556
0.4449	11.0	33	0.4449	5.5556
0.4353	12.0	36	0.4344	5.5556
0.4257	13.0	39	0.4243	5.5556
0.4168	14.0	42	0.4147	5.5556
0.4075	15.0	45	0.4053	5.5556
0.3985	16.0	48	0.3960	5.5556
0.3893	17.0	51	0.3866	5.5556
0.3801	18.0	54	0.3770	5.5556
0.3706	19.0	57	0.3673	5.5556
0.3609	20.0	60	0.3574	5.5556

Framework versions

• Transformers 4.49.0
• Pytorch 2.2.1
• Datasets 3.3.2
• Tokenizers 0.21.0